The Probabilistic Movement Primitives (ProMPs) is an essential issue and framework for robotics Learning from Demonstration (LfD). It has been successfully applied to the robotics field in tasks such as skill acquisition and Human-Robot Collaboration (HRC). In this paper, we focus on its adaptability in the HRC scenario, in which the adaptability of the ProMPs allows the robot to predict the future movement of its human partner and plan its movement accordingly, given the observed human movement. Most of the existing works about the application of the ProMPs in HRC either only focus on the estimation of the weights on-line and lack the estimation of the phase parameter or merely depend on the prior distribution of the phase parameter. As a result, these methods can lead to a misinterpretation of the basis matrix when the divergence between the prior distribution and the posterior distribution of the phase parameter becomes large, resulting in a divergence of the estimation of the weights. In this paper, we propose a Dual-Filtering method for the ProMPs, which is able to simultaneously on-line estimate the weights and phase parameter for the ProMPs. The preliminary experimental result demonstrates the proposed method is able to provide better prediction performance and more accurate estimation of the phase parameter in comparison with the previous works.

Household chores service is one of the desirable functions for a service robot. Object classification is the most important function when searching for objects. We consider that the major concern is that the robot should not misclassify the object. If the robot misclassifies household object images, it will then perform the household tasks with the wrong object. This may cause serious damages to a service robot and to the user. This concept gives us the insight that the precision of the classification must be very high by setting a confidence threshold so that it then can claim a reliable service robot. By exploring this concept, we develop a more convincing indicator, Classification Reliability, to reveal the reliability of deep learning model. Moreover, we develop a fine-tune rule base to continuously regenerate more proper training dataset for the CNN model to increase reliability. Experimental results demonstrate that the CNN model fine-tuned by our closed-loop system achieves the reliability which is higher than the other similar effects such as DenseNet on the CIFAR-10 dataset.

With recent development of robotic technology, it is increasingly common that robot coexist with human, in which humans and robots share a common workspace and work in close proximity. To maintain efficiency and ensure safety under these circumstances, robot should have the ability to predict the future human motion based on the observed on-going motion. In this paper, we present a methodology for on-line inference of human intention and prediction of human hand motion. The proposed framework is built using Probabilistic Dynamic Movement Primitive (PDMP). In the off-line stage, a set of PDMPs is constructed based on the recorded demonstrations and they will then be used for inferring human intention and predicting human hand motion in the on-line stage. A proof of concept evaluation is carried out in a tabletop manipulation task. Experimental result shows the proposed framework achieve high performance in human intention inference and in the trajectory similarity between the predicted and the actual hand movement under the normally defined environment. We also show the proposed framework can adapt and generalize to the newly defined environment.

Indoor localization is one of the most important technologies for a mobile robot working in indoor environments. Laser-based robot localization methods provide good performance but is still problematic for kidnapping and re-localization issues. Monte Carlo Localization (MCL) has been introduced as a useful method that can deal with kidnapping and re-localization problems. However, MCL requires long computational time when the robot is working in a large scale environment. In this paper, we propose indoor re-localization solutions based on hierarchical Wi-Fi fingerprinting for an autonomous mobile robot. Hierarchical Wi-Fi fingerprinting requires less time for re-localizing the robot compared to laser-based MCL methods. Hierarchical Wi-Fi fingerprinting can immediately provide information to approximately localize which room the robot is in and the robot's approximate position after robot loses its position. The room-level localization is achieved by dendogram-based support vector machines (DSVM) and the robot's approximate position is estimated by Wi-Fi fingerprinting. After the robot's approximate position is calculated, more accurate robot position is attained by performing feature matching between laser measurements with the local 2D map. Finally, we propose the experiments in a practical indoor space. We use the laser-based generated SLAM map using adaptive Monte Carlo localization to compare with the proposed method. The experimental results show that using hierarchical Wi-Fi fingerprinting with laser-based localization can have good performance on robot re-localization.

This paper presents a face portrait with cartoon stylization painting and associated algorithms with the visual feedback system to paint like a human cartoonist. The robot cartoonist creates the artwork in two stages-cartoon style transformation and robot artist for colorful painting. In the cartoon style transformation stage, it transfers human portrait photos to cartoon style by face detection and alignment, which can effectively decompose the face into individual components then replace by cartoon facial components. In the second stage, the robot uses an eye-in-hand system to obtain five basic colors (cyan, magenta, yellow, white and black) to automatically mix a variety of colors automatically. For painting strategy, we start with the outline of the face, which we use non-photorealistic rendering (NPR) to generate hand-painted strokes. After that, the robot artist will implement painting the facial features. We also demonstrate the success of this proposed research.

Due to the widespread use of industrial robots in market, its application has extended to welding, painting, and freight handling. And tool coordinate calibration is regularly modified after tool replacement due to collision accident or routine maintenance. After tool replacement, operators often rebuild tool coordinates. This is the traditional mode of operation in the current industrial practices. However, smart factory will make artificial intelligence method replace manual method. This paper presents a system independent method for automatic calibration of the tool coordinate system which is faster, simpler, cheaper and more effective than the manual method. The proposed method required images to be captured using two “eye to hand” cameras and one “eye in hand” camera. Tool position data is then acquired through CamShift and MeanShift algorithm for image trajectory tracking along with coordinate system conversion, several methods like PCA, LDA can deal with the vision data. Optimal Deep Neural Network (DNN) method error compensation of a robot allows the tool to automatically run with the calibration system functions. We have developed a 6 degrees of freedom(DoF) industrial robot for this experiment. Nine different kinds of DNN models are built and finally with suitable tool coordinate error compensation for the current robot, tool calibration can be achieved adaptively and efficiently.

Diagnosis and prediction of the health status of vehicle components production line machine is the core requirement for global manufacturing system. With the development of Internet of things (IoT), there are enormous big data of production line could be collected quickly and stored in large quantities. The development of artificial intelligence makes it possible to deal with big data efficiently. Due to the industrial requirement of health self-diagnosis for vehicle production line, this paper presents a method based on Artificial Neural Network (ANN) to diagnose the health status of production line machines using the data produced by the machines. The PID control parameters of motors are segmented to simulate the health status of the machines in a long duration. We use three kinds of artificial neural network (ANN) methods to train the model of the relationship between the large data trend and the diagnostic score of the machine, it is demonstrated that it becomes more efficient than traditional empirical analysis to improve the speed and accuracy for diagnostic and prediction of machines health status.

The objective of this paper is to present a percussive massage based on robotic tapping motion, and it is also the first paper to discuss the massage technique for this kind. The tapping motion is different from the common contact motion, it satisfies with the short-time contact on human muscles, and the robot end-effector itself has an initial velocity and acceleration to make an impulse contact. After contacting with human muscles, the robot manipulator stops and end-effector repeatedly performs the tapping motion. To tackle with the tapping motion problems, we utilize the online trajectory generator (OTG) based on the impedance control, which ensures the safety of human without any possible accident which can hurt human. We implement the tapping motion with a dual arm robot developed at the iCeiRA lab in NTU. By the Cartesian space teach function and the concept of virtual point, the tapping force and contact position can be changed adaptively to meet different needs.

For industrial robotic applications, conveyor tracking is one of fundamental function in the robot manipulator. While the target is moving on the production line, the task becomes sophisticated problems. Thus, a distinct grasping method under visual control system is definitely one of the essential solutions. In this paper, we propose a tracking strategy on moving objects for a robot arm object fetching system combined with distinct recognition algorithm. In addition, the grasping pose of robot arm is corrected by visual feedback system. The system is separated into two parts and discussed in detail. Each part owns its core algorithm to complete industrial tracking and fetching tasks. Because of limitation from the environment, the working conditions will also be illustrated. Eye to hand and eye in hand both contribute to the visual feedback system. Grasping pose for each type of workpiece is adjusted by tracking and optimization algorithms. The result of object recognition is enhanced by visual system in determined pose and orientation. All experimental results were completed by a 7-DoFs robot arm developed in our lab at NTU.

In this research, we aim to implement the ability to navigate through unknown environments according to some given instructions on mobile robots. We proposed a Recursive Neural Network model, which takes the user instructions and data acquired by laser range finder as input, and outputs control velocities to make the mobile robot navigate in an unknown indoor environment. Instructions are using commonly verbal expressions, and instead of establishing models to represent the semantic meanings, we use the concept of word vectors to feed the information directly into the neural network models. Using human-controlled navigating records as training data, robots learn how to execute navigations according to various of different instructions. Experiments were conducted under both simulation and real world environments, and results show that the robot can successfully navigate to the goal positions without having any prior knowledge about the environments.

As humanoid robots start entering to the human beings environments, the collision between the robot and other objects is inevitable. To solve this problem, we propose the human mimicking push-recovery strategy. With the arm mechanism, the humanoid robot utilizes human-like strategies to deal with the unexpected collision. We analyse the human push-recovery movement to the external force. Thus we can design the trajectory generator to mimic human beings reactions. To reduce the modelling error and improve zmp stability, we utilizes the five-mass with angular momentum model as humanoid robot model. We design the safe bound examination and the Center of Mass (COM) state estimator to judge the collision stage. The safe bound examination is used to guarantee the walking stability. Moreover, the COM state estimator is used to determine whether the pushing is over or not. The push-recovery control system is implemented on the humanoid robot developed in our NTU-iCeiRA lab. The purpose of this paper aims to integrate different push-recovery strategies while encountering different collision conditions based on the five-mass with angular momentum model.

This paper proposes a robotics therapeutic massage evaluation system using electroencephalogram (EEG) signal analysis approach. An anthropomorphic dual arm robot is developed in this work for massage application. Compare to other massage systems, our dual arm robot can provides diversified massage techniques through the impedance control of both Cartesian space and joint space. In order to evaluate the effectiveness of robotic therapeutic massage, the experiments are conducted by recording EEG signals of subject before and after robotic massage. Independent Components Analysis (ICA) is used to filter out artifacts of EEG signals. After signal processing, the power of delta, alpha and beta rhythms are analyzed. The experimental results show an increase in delta power and a decrease in alpha power which represents the relaxation response of subjects. These results give the scientific proof of the effectiveness of robotic therapeutic massage.

This paper presents a robot capable of painting colorful pictures with a visual control system like human artists. It can use only five basic colors (cyan, magenta, yellow, white and black) to mix a variety of colors. After receiving a picture, the Robot Artist creates the artwork in two stages-the underpainting and the refinement. In the underpainting stage, it covers the canvas with a thin layer of acrylic paint, which can effectively set the basic tone of the painting. In the second stage, the robot uses the camera attached to its wrist to obtain the current artwork and refine it repeatedly. During the refinement, the methodology of non-photorealistic rendering (NPR) is also introduced to generate hand-painted strokes, which will be painted on the canvas to refine the artworks.

Since social, assistive and companion robots need to navigate within human crowds, understanding spatial social conventions while designing navigation solutions for such robots is an essential issue. This work presents analysis of an socially compliant robot motion strategy that could be employed by social robots such as humanoids, service robots or intelligent wheelchairs, for approaching and joining humans groups in interaction, and then become an equitable part of the interaction. Following our previous work that formalized the motion strategy, a detailed synthesis is presented here with experiments that validate the proposed system in the real world.

In this paper, a service-oriented multiagent system (SoMAS) for the control and analysis of the cyber–physical system (CPS) in manufacturing automation utilizing a noncontact dynamic obstacle avoidance seven-DoF robot arm is presented. The interfaces of the services which the robot arm subsystem should provide to fully exploit its capability are identified. Specifically, the services of moving, object recognition, object fetching, and safety of human–robot interaction are considered as the fundamental functionalities that the robot arm should provide. The way to evaluate the quality of services (QoS) for the robot arm subsystem is also explained. To build such a robot arm subsystem, the system architecture is proposed. Also, implementation for the subsystem which includes: 3-D model-based object recognition, grasp database for object fetching, and online noncontact obstacle avoidance for the safety of human–robot interaction is provided. The experimental results demonstrate that the capability of 3-D model-based object recognition, object fetching, and dynamic collision avoidance are successfully implemented.

This paper presents the algorithm of human pose estimation in 3-D space using adaptive control law with point-cloud-based limb regression approach. The proposed approach is a data-driven method for 3-D pose estimation of human. In addition, we exploit the inverse relationship between the estimated parameter and the pose of limb, proposing a hybrid scheme, the combination of indirect adaptive scheme in Cartesian coordinate, and indirect adaptive scheme in spherical space. The experimental results show the ability of error tolerance, even dealing with sparse, partial, and noisy limb data. The comparison of the ground truth is provided with the standard dataset. The computation speed is sufficient, as it is expected to be used in real-time applications or as the 3-D feature in other recognition frameworks. Finally, the proposed algorithm is demonstrated with the point cloud sensing in real scene and the experimental results are included in this paper.

The objective of this paper is to present the model-based 3D object recognition and fetching by a 7-DoF robot with online obstacle avoidance for factory automation. The robot can fetch the random type of 3D objects with arbitrary pose using a 3D visual camera. Object recognition pipeline based on different descriptors are compared in order to decide the best choice for this scenario. A correct grasp is taught by human operator using programming by touch. As a result, the time of configuring and setting up the robot drastically decreases and thus saves a large amount of preparation time before the robot can actually get to work. Considering the safety issue of the human operators that cooperate with the robot, as well as the uncertainty of the environment such that a moving obstacle might present, or the environment might change for some reason, an online obstacle avoidance algorithm is also integrated into the system. Experimental proof of principles has been successfully implemented.

Bipedal robot has many advantages to locomotion in complex environments that wheeled robot cannot achieve. However, because of biped robot's complicated mechanism and the naturally unstable system, it is very susceptible to disturbance from the environment and humans in comparison with wheeled robots. Therefore, the prime task for biped robot walking is to maintain dynamic balance and recover from the perturbation when it occurs unexpectedly. In our previous work, the strategy in the sagittal plane has been studied. While the lateral one has not been considered yet. The objective of this paper is to deal with perturbation in lateral plane, and combine them to a planar direction strategy. Various recovery approaches are utilized to stabilize the biped robot while walking. The methods take advantage of angular momentum by quickly swinging leg or rotating trunk. Also, it is determined in the way as human beings encounter pushing force. The proposed method has been successfully implemented on the biped robot developed in our iCeiRA laboratory.

Mecanum wheels can apply a flexible mobility to a mobile robot, in this paper, a mobile manipulation robot is designed with combination of a Mecanum wheeled mobile platform and a 7-degrees of freedom (DoF) robot manipulator. Due to the mechanism of Omni-directional mobility, the Mecanum wheeled robot suffers with slip errors on the free rollers, these errors may be reduced with additional sensors and appropriate feedback control. However, while the mobile platform and the manipulator operate simultaneously, the displacement of the center of gravity of the whole robot increases the complexity to the motion control. To solve this problem, we analyze and model the effect of inconsistency of slippage errors caused by the unbalance of normal force on the wheels, and propose an on-line adaption slippage model for the Mecanum vehicle to generate a better control system to reduce these errors. We proof the effectiveness by using our proposed adaption model and on-line compensation control with experiments.

In this paper, we propose a modularized architecture for a robot arm object fetching system integrated with 3D CAD-model based object recognition system that can cope with objects in random types and poses. The interface and the core functionality of each module in the architecture are discussed in detail. Implementation of each module is also conducted. Furthermore, the assumptions and the working conditions behind each module are carefully examined. We develop the system based on our previous work and enhance the recognition module. To proof the feasibility of our architecture, 3D object recognition and fetching demonstration are successfully implemented, and the result of object recognition, teaching by touching, and fast grasp synthesis are successfully demonstrated.

This paper aims to develop a transparent object recognition algorithm which can integrate with visual cues of transparent objects to enhance the retrieval result. Due to the special characteristics of transparent objects, very few vision methods exist to identify them. To achieve our goal, we first use active depth sensor combining with image pre-processing techniques to retrieve transparent candidates. We then propose a transparent candidate classification algorithm using visual cues in color image to further distinguish the transparent ones out of candidates. Experimental results show that our algorithm can achieve good transparent object retrieval results and outperform methods based on depth image. In addition, this algorithm can be integrated with pose estimation methods on robotic bio-laboratory automation applications.

In current flexible endoscopy, manipulating the endoscope is a major problem for surgeons. No surgeon can master the procedure without much experience. Accordingly, the objective of this study is to refine our Robotic Flexible Laparoscope System (RFLS) with more useful functions. We design a new kind of laparoscope mechanism suited for our RFLS. Furthermore, in maneuvering the laparoscope intuitively, the most significant advantage is the ability of RFLS can at least save three anatomical positions, which could be retrieved by a single command with small errors. Moreover, protective algorithm is also included in our system to make sure that over rotation of the gyro sensor, which is mounted the head, does not cause any damage. Experimental results demonstrate that the articulated flexible laparoscope can always memorize the required points and return back to those points in all necessary orientations.

In this paper, we propose a leg compliance method and strategies for ZMP-based bipedal walking robot on continuous uneven terrain without gyro sensor by combining soft joint, impedance, force and position control. We divide biped robot into support leg and swing leg to apply different control methods, and distinguish period of swing leg into four procedures, which adopt different strategies. By this way, robot achieves walking on uneven slope. Besides, when robot walks on normal flat ground, the proposed compliant landing method can also decrease the impact of ground reaction force (GRF) and reinforce stability during transition from single support phase (SSP) to double support phase (DSP). The validity of the proposed approach and strategies are verified and achieved with experiment biped robot in our lab.

Gender roles influence behavior in social interactions. Thus, real-time gender recognition is essential in Human-Robot Interaction (HRI) for providing timely gender information to improve the experience of HRI. Considering the HRI scenario, a 3D-human-body-shape-based gender recognition is investigated. The 3D information is obtained by processing the depth image from an RGB-D camera. In addition, a machine learning method based on a Support Vector Machine (SVM) was applied. The experimental results showed that our system could achieve real-time accurate gender recognition. It enriched the diversity of existing methods for HRI application.

Intelligent robot for improving the quality of human life is desirable. To let robot perform the intelligent functions of massage which can serve for humans, adaptive massage trajectory generator is needed to suit for different body shape with different people. Base on the state of the art human pose recognition techniques, we develop a methodology to build the relationship between camera and human body and generate the specified massage trajectories. Assume we can detect human pose and label each part of human body, we use the RANSAC algorithm to estimate the frontal and sagittal planes and refine it by calculating the minimum moment of inertia of point cloud. The experimental results demonstrate our work which provide a useful method for massage trajectory generation. As an example of service robot application, this method is useful for executing the massage application autonomously.

In this paper, an erect stair climbing robot, which has been able to ascend stairs, is further developed for descending stairs automatically. The somatosensory control functions are also implemented for a human rider to drive the robot on flat ground as a personal vehicle instinctively. In order to face front whether climbing up or down stairs, the robot is designed with proper mechanical structure and mass distribution. Moreover, the perception and control methods of descending stairs are modified from the methodologies of ascending stairs in our previous works so that the robot can handle both ascending and descending stairs effectively under similar principles. As for the somatosensory control part, the functions of controlling the robot to move back and forth using the rider's body motion is realized, which has more advantages than Segway PT since it can stand still when the rider wants the platform to stop moving, and the robot motion under the control of the human rider is stable enough for indoor applications. In the experiments, the stair-climbing abilities and somatosensory control functions are both successfully demonstrated, which proves that the design and implementation of the robot system are feasible and efficient. We are looking forward for more functions and applications of the robot, including our ultimate goal of carrying a human rider to climb stairs in an erect way with somatosensory control simultaneously.

The objective of this paper is to propose an approach for generating on-line walking patterns for a biped robot to navigate towards destination and reactively avoid obstacles. Kinect-based sensor system detects obstacles with velocity and dimension estimated for obstacle modeling. With identified obstacles, repulsion vectors of potential field given by obstacles modify walking path of a biped robot. On-line path generation (OPG) is proposed to generate walking path towards a target configuration including position and orientation. With reference mapping combined with walking pattern generator based on zero momentum point (ZMP), reactive walking patterns free of obstacle collision are thus created. The effectiveness of this method is confirmed on a biped robot developed in our lab, and the robot is capable of avoiding obstacles while maintaining ZMP stability. The contribution of this paper is to propose OPG for obstacle avoidance of biped robots by extending potential field method. The Demonstration of the proof of concept has been successfully conducted.

The objective of this paper is to develop conveyer object tracking and fetching system. The traditional visual servo system for parallel robot is based on Eye-to-hand method which suffers from the narrow field of view (FOV). The basic step of moving assembling is multiple objects tracking which needs broad work space and big FOV of the parallel robot. In this paper, a Hybrid Eye-to-hand and Eye-in-hand Visual Servo System for parallel robot is proposed. Via this hybrid system, the parallel robot can achieve (multiple) object tracking and fetching with priority planning we designed. The eye-to-hand visual system with low resolution camera provides the estimations of the velocities and poses of the workpieces. Also the fetching priority is planned with this system. All aforementioned information is transferred to the eye-in-hand visual servo system for further use. Once the object appeared in the FOV of eye-in-hand system in the estimated position, the high resolution camera of the system is employed to acquire the pose information of the object with high accuracy and track it frame by frame. The hybrid system is successfully demonstrated and implemented on parallel robot which not only broadens the working space of parallel robot but also achieves great efficiency and accuracy.

The objective of this paper is to develop an analytical inverse kinematic solution for a 7-DoF redundant manipulator with offsets at shoulder and wrist. Firstly, the key ingredient is to derive closed-form inverse kinematic solution using different joint as redundancy based on parameterization method. By having these redundant parameters,we obtain different null-space self-motion pattern of the arm. We then can select a proper redundant joint. The method is consistent with calculating augmented Jacobian matrix. Lastly, the derived equations are verified with a modularized 7-DoF manipulator with offsets at shoulder and wrist, which consists of modularized components designed and built by our laboratory.

This paper describes Cartesian position and force control with adaptive impedance/compliance capabilities for a humanoid robot arm including velocity control incorporate with End-Effector Fixation Control (EEFC). We successfully demonstrated the end-effector positioning to reach the target point in Cartesian coordinate frame based on force control without solving the inverse kinematic problem. The advantage of this function is obvious in moving the end-effector to go to the target point safely because the robot arm can be stopped or disturbed during the movement. The impedance gain can be adjusted adaptively according to the moving trajectory conditions. As the result, the end-effector can reach the presetting target point. We also have experimentally demonstrated the effect of adaptive impedance/compliance control, velocity control and EEFC which are useful in many applications, such as laparoscopic surgery, industrial taper cutting compensation and feeding food to the impaired person.

There are many modern applications require the function of age classification. In this study, we propose a method to classify human age using appearance images and apply it to the human-robot interactions. We first confirm that facial features based on craniology are not discriminative under the condition of seven age-groups classification. Next, our system is designed to have two stages. One is image preprocess stage; faces are detected using Haar-like features with Adaboost algorithm. Our image database is from FG-NET and MORPH databases so that we have high degree of complexity and difficulty in recognition. Then images are trained by support vector machines (SVM). To have higher recognition rate, we train RBF (radial basis function) and linear kernel models at the same time, and decide the final results by comparing the two models. These improve the accuracy under age of 30 to 49 years old while the linearity is preserved under age of 0 to 29 and above 50 years old. The final age recognition rates achieve 91.5% for female and 96% for male. We also compare the age-group classification results with subjective questionnaires, and it demonstrates that the proposed system has better performance than human's subjective estimation.

The objective of this paper is to develop a method of walking pattern generation with non-constant body height towards quasi natural walking. Unlike the traditional LIPM or GCIPM based method, our proposed method divides the walking cycle into 3 parts, including lowering down, rising up and inverted pendulum process, all described by specific trajectory equations. After modelling the dynamics of the biped walking robot with three mass points and properly designing ZMP reference, smooth spatial hip trajectory with non-constant height is obtained, as well as joint trajectories solved by inverse kinematics. The stability of biped walking pattern is verified through simulation and actual walking experiments on a biped robot developed in our lab.

In this paper, an automatic erect stair climbing mobile robot is developed for indoor service applications. The design of high center of mass, tilt axis near ground, and the triangular wheel-legged structure enable the robot to climb stairs in a dynamic and self-balancing way. The wheel-legged mechanism also keeps the advantages of differential-wheeled mobile platforms when moving on flat ground, such as easy to control, saving power, and zero turning radius. Moreover, the overall mechanical design fits the requirements for a service robot to have proper height and small footprint, which can make human-robot interaction more natural and comfortable in indoor environment. Since the perceptual and control system of the robot are both well integrated, we successfully demonstrate the stair-climbing function and prove that the design and implementation of our work are feasible and efficient. As a result, the robot is expected to be a prototype of universal platform for indoor mobile service robots in the future.

Indoor localization and monitoring system of robots and people are essential issues in robotics research. Several monitoring systems are currently under development by different investigators but they do encounter significant difficulties. For instance, a Pyroelectric Infrared (PIR) system provides less accurate information of human location and is restricted when there are multiple targets. Furthermore, a Radio Frequency (RF) localization system is constrained by its limited accuracy.

It is desirable that an Intelligent Service Robot (ISR) will not only construct the environment map but also recognize the meaningful symbols/signs in the building it services simultaneously. The objective of this paper is to describe an ISR autonomously estimate the environment structure and simultaneously detect the commonly recognized symbols/signs in the building. The result is an information map constructed by the environment geometry from a laser range finder and the indoor indicators from visual image. To implement this indicative information map, sensory fusion techniques: batch Maximum Likelihood Estimator (MLE) and Covariance Union (CU) are tactically utilized for robust pose and sign estimations in a single SLAM process. Also, a 2.5D indicative environment map has been constructed rapidly with the 3D Mesa SwissRanger. We have successfully demonstrated the proof of concept experimentally and summarized in the concluding remarks.

In this paper we introduce a combining method with multi-sensor fusion for hand sign recognition and apply this method on service robot. Hand sign recognition is an essential way for Human-Robot Interaction (HRI). Sign language is the most intuitive and direct way to communicate with impaired or disabled people. Through the hand or body gestures, the disabled can efficiently let caregiver or robot know what message and instruction they want to convey. In this paper, we propose a combined hands gesture recognition algorithm which combines two distinct recognizers. These two recognizers collectively determine the hand's sign via a process called combinatorial approach recognizer (CAR) equation. These two recognizers are aimed to complement the ability of discrimination. To achieve this goal, one recognizer recognizes hand gesture by hand skeleton (HSR), and the other recognizer is based on support vector machines (SVM). In addition, the corresponding classifiers of SVM are trained using different features which are Gabor feature and raw data. Furthermore, the trained images are using Bosphorus Hand Database and in addition to taking by us. A set of rules including recognizer switching and combinatorial approach recognizer (CAR) equation is devised to synthesize the distinctive methods. To achieve the human hand sign recognition in the correct statement, we equip two sensors on the service robot and implement the above algorithms on it. These two sensors are CCD sensor and laser range finder (LRF). We have successfully demonstrated signs recognition experimentally with quite satisfactory proof of concept results.

Emotional interaction with human beings is desirable for robots. In this study, we propose an integrated system which has ability to track multiple people at the same time, to recognize their facial expressions, and to identify social atmosphere. Consequently, robots can easily recognize facial expression, emotion variations of different people, and can respond properly. In our facial expression recognition scheme, we fuse Feature Vectors based Approach (FVA) and Differential-Active Appearance Model Features based Approach (DAFA) to obtain not only apposite positions of feature points, but also more information about texture and appearance. With the obtained useful information, FVA can classify the emotions according to comparison with the distances and ratios of feature points, and DAFA can distinguish emotions from classical machine learning on a low dimensional manifold space. Furthermore, emotion recognition of multiple people at the same time is extended. Based on the proposed algorithm, multiple person emotion analysis and social atmosphere identification can be achieved, which makes the relationship between people and robots much closer. Experimental results demonstrate that the proposed algorithms can recognize facial expressions accurately and robustly. The ambient atmosphere identification system is implemented with a young Einstein robot head in our laboratory.

The objective of this paper is to review the theories and approaches of multisensor fusion and integration (MFI) with its application in mechatronics. MFI helps the system perceiving changes of the environment and monitoring the system itself. Since each individual sensor has its own inherent defects and limitations, MFI merges the redundant information acquired by multiple sensors synergistically to provide a more accurate perception and make an optimal decision in further. The wide application spectrum of MFI in mechatronics includes the industrial automation, the development of intelligent robots, military applications, biomedical applications, etc. In this paper, the architecture and algorithms of MFI are reviewed, and some implementation examples in industrial automation and robotic applications are presented. Furthermore, sensor fusion methods at different levels, namely, estimation methods, classification methods and inference methods, the most frequently used algorithms in previous researches with their advantages and limitations are summarized. Applications of MFI in robotics and mechatronics are discussed. Future perspectives of MFI deployment are included in the concluding remarks.

The multi-finger robot hand (IRA-Hand I) is specially designed for massage applications. Surface electromyographic (SEMG) evaluation of therapeutic massage effects using multi-finger robot hand are presented in this paper. After an isometric 50% MVC (maximum voluntary contraction) of human back muscles is performed for 90 seconds, SEMG signals of the subject show a muscular fatigue. A grasp-kneading massage by the robot hand is applied on the subject's shoulder for recovering from muscular fatigue. To evaluate the therapeutic effects of massage, the SEMG signals measured from the trapezius muscles before and after the massage therapy are analyzed. Electrical activity (EA) and median frequency (MF) of the SEMG signals are calculated as indexes of the muscle physiological states. The experimental results show that EA increases from the occurrence of fatigue, while MF shifts towards lower frequency in the spectral distribution. After massage, a decrease in EA and an increase in MF are observed which demonstrate the effectiveness of recovery through the grasp-kneading massage by the robot hand. In addition, the joint analysis of EMG spectrum and amplitude (JASA), which considers the changes in time domain and in frequency domain simultaneously, verifies that the therapeutic massage recovers the trapezius muscle from fatigue effectively. For comparison, the experiments with a massage specialist performing the massage therapy are conducted with the same procedures. It is evidenced that the robot hand massage has even better effectiveness than the human hand in most cases.

Indoor monitoring system for robot and human is an important issue in robotics research. In this study, we propose an indoor monitoring system based on wireless and pyroelectric infrared (WPIR) sensory fusion system. Pyroelectric infrared (PIR) system alone for localization provides less accurate information of human location and suffers from multiple targets localization. Furthermore, radio frequency (RF) localization system alone is also constrained by its limited accuracy. We develop a sensor network based localization method called WPIR inference algorithm. This algorithm determine the fused position from both PIR localization system and radio frequency signal localization system which utilizes received signal strength (RSS) model. We have developed and experimentally demonstrated a WPIR sensory fusion system which can be applied in localizing targets, such as human and robot. With the proper localization mechanism in the indoor environment, the provision of appropriate services for people can be realized with less expensive cost.

In this paper, we propose a hybrid visual tracking algorithm that combines two discriminative trackers. Discriminative trackers treat tracking as a classification problem, that is, they try to distinguish targets from backgrounds, and usually the trackers incorporate classifiers. The two trackers collectively determine the new object location via a process called confidence fusion. The two trackers are aimed to complement the ability of discrimination. To achieve this goal, one tracker extracts image features pixel by pixel, and the other extracts image features over several rectangular regions. In addition, the corresponding classifiers are trained using different learning algorithms. We not only model object tracking as a binary classification problem but also model it as a three-class classification problem. On-line learning algorithms are used to update the classifiers during tracking so that the trackers are adaptive to the variations of the appearance of the target. A set of rules including tracker switching and confidence fusion is devised to synthesize the two trackers. The experimental results show that our approach is competitive with other popular tracking algorithms.

The IEEE/RSJ 2010 International Conference on Intelligent Robots and Systems (IROS 2010) will be held on October 18–22, 2010 at the Taipei International Conference Center (TICC), Taipei, Taiwan. IROS 2010 is the 23rd year of the conference, since its inception in 1988. The conference theme is “Intelligent Robotics in the Next Transition” The theme of IROS 2010 is chosen such that the conference would continue reflect the ever increasing interest in interactions, co-works, and co-existence of robots with human, cognitive robotics and other forms of intelligent machines and systems, with the goal for improving the better quality of human life. IROS 2010 continues to attract a large number of participants, paper submissions, and accepted papers whose quality remains high. IROS 2010 had 1798 papers submitted from more than 50 countries. We will have 828 papers for oral presentation and 144 for interactive discussions at the conference after first going through a rigorous review process, then by second going through a quality control process conducted at the Program Committee (PC) meeting in May 2010. The technical Program of IROS 2010 consists of 1 Keynote Talk, 3 Plenary Talks, one day Robot Industry Forum, half day Robotics Education Forum, 140 technical sessions in 14 tracks, 17 workshops/tutorials, and 16 video presentations, 14 exhibitions, and Robot Competitions.

This paper presents an electromyographic (EMG) signal integrated multi-finger robot hand control for massage applications. This research explores the feasible application of multi-finger robot hands except for the use as prostheses and grasping applications. The forearm EMG of a person who is massaged by the human hands is recorded and analyzed statistically. First, the root mean square (RMS) of the raw data is computed as the discrimination between normal and contracted states of the muscle. Then the EMG signal at contracted state is further divided into painful and comfortable groups based on the impulse factor which is defined to estimate the sharpness of waveform variations. As a consequence, two discriminative values of the EMG signal are generated to distinguish painful and comfortable feelings. Based on the relationship between the human feeling and the massage force, we get an appropriate range of input commands of the robot hand for massage applications. A grasp-kneading massage is performed on the human shoulder to verify the proposed process. As a result, an effective and comfortable massage using the multi-finger robot hand is realized.

Indoor human monitoring is important for providing various services for individuals. In this research, we propose an indoor monitoring system based on wireless and pyroelectric infrared (WPIR) sensory fusion system which can be embedded and integrated with the traditional fire/smoke detector where they are usually installed on the ceiling by the use of pyroelectric motion sensor and low power wireless communication device. Basically, each of radio frequency (RF) localization system and pyroelectric infrared (PIR) monitoring system provides the coarse information of individuals' location respectively. We develop a sensor network based localization method called WPIR intersection algorithm to determine the fused position from PIR and radio frequency signal localization system which utilizes received signal strength (RSS) model. We have developed and experimentally demonstrated a wireless pyroelectric sensory fusion system and WPIR intersection algorithm which can be applied in multi target tracking issues against pure PIR sensing system. Furthermore, WPIR intersection algorithm provides higher accuracy than PIR or RF localization respectively. With the proper localization and tracking in the indoor environment, the provisions of appropriate services for people can be realized.

The objectives of this paper is to develop a system for an autonomous mobile robot for searching and tracking sound source based target. The target can be a power charge docking station. Finding objects or events by following a sound source direction is an intuitive response for human and animal when they cannot see the target. However, there are relative fewer discussions about sound source tracking in robotics field. We design a navigation algorithm integrated with time delay of arrival (TDOA) techniques based on cross-correlation. To confirm the practicality of our approach, we design a robot docking scenario and develop an autonomous mobile robot with two microphone pairs, a camera and a laser range finder to conduct the experiments. According to the experimental results, our approach reduces the complex localization problems to save computing power, and have an acceptable result in an indoor environment.

Intelligent Service Robot (ISR) has become increasingly noticed because of new devices and novel technologies that make ISR a truly handy human aid in areas like medical care, security patrol, tour guide and edutainment. Therefore, how to provide an applicable map for ISR to autonomously navigate inside a building for task execution becomes an imminent issue. This paper investigates an information enriched map constructed by the environment geometry from laser range finder and the indoor directive signs commonly seen in living/working environment from camera image. To implement this enriched map, multi-sensor fusion techniques are utilized for robust pose and sign estimations. Furthermore, an improved alignment technique is applied to reduce the computational complexity in a single Graph-SLAM process.

Within an intelligent building, it is expected to offer various intelligent services by recognizing residents along with their lifestyle and needs. One of the key issues for realizing the intelligent building is how to detect the locations of residents, so that it can provide the interactive services based on the identified need and requests. In this work, we develop a wireless pyroelectric sensory system embedded with traditional fire detector, which can be implanted on the ceiling. Both of wireless transmission model and pyroelectric sensor monitoring system can provide the rough information of residents' location respectively. These data can be further improved by reducing the sensory uncertainty through covariance intersection (CI) data fusion method. Pyroelectric localization system suffers from multi target tracking and wireless pyroelectric sensor system works this issue. With the location obtained from wireless pyroelectric sensor system, intelligent building can offer suitable services.

The power management system for autonomous mobile robots is an important issue for keeping robots in their long-time functionality. Recharging is necessary before the power of the robot has exhausted. In this paper, we propose a multi-sensor fusing method using intensity and range data fusion with covariance intersection approach to locate the robot pose while performing the docking for recharging. An artificial landmark is employed as a visual cue on a docking station in order to recognize the location by using inverse perspective projection. At the same time, the range data acquired by laser range finder are modeled as multiple line segments which are the hypothetical walls in the environment. Then the geometrical relationship between the robot and the docking station is estimated much more precise by using covariance intersection approach. We have demonstrated the success of the proposed algorithms through experimental results.

In this paper, we have constructed an artificial ecological system (AES) used to manage sensor's behaviour and maintain sensor network. We have defined two kinds of species for sensor nodes, one is called prey and the other is called predator. It is desired to keep these sensor nodes reach the blanket coverage and maximize, two species exist and ensure they will not die out. We use two methods namely self-deployment and self-maintenance. The sensor nodes dynamics model (SNDM) based on the molecular dynamics is used to implement the diffusion, and the sensor nodes ecological model (SNEM) is used to maintain the sensor nodes. With these two models, we can create an ecological balance environment with automatic recharge, automatic recycle and automatic quantity control.

In recent years, rapid tooling (RT) has become increasingly important because of the requirement of rapid manufacturing. RT can be used to produce temporary molds as well as permanent molds in mass production. RT provides more benefits from the design stage to mass production. The objective of this research is to develop the direct metallic RT system and identify process parameters of laser cladding. A 2.5-kW Nd-AG laser system, a metallic powder feeder system, a lateral powder nozzle system, a PC-based motion controller, and a shielding gas-supply system are incorporated for the hardware architecture in the proposed RT system. The critical process parameters influencing the dimensional accuracy were analyzed with Taguchi's experimental design. From the analysis of variance computational results, the factors "Scanning speed" and "Tool path offset" are the influential factors affecting the layer thickness of RT parts. From the experimental results, the dimensional accuracy of the dual nozzle RT system is better than the single-nozzle RT system. Finally, the proposed RT system could fabricate the mold inserts. After finishing the mold inserts and assembling the plenary injection mold, the injection molding machine can generate acrylonitrile butadiene styrene plastic parts with excellent quality. Note to Practitioners-Rapid prototyping (RP) technologies have provided a new way of making models or producing visualization prototype in a fast fashion. However, there is an urgent demand from the industry for much faster ways to make physical prototypes, with the desired material using the appropriate production method. RP technologies have introduced a new generation of RT processes. Producing tooling directly from computer-aided-design (CAD) models is regarded as an important process of reducing the cost and time to market for new products. In this paper, a direct metallic rapid tooling (RT) system is developed to fabricate the metallic parts and molds directly from 3-D CAD models. The experimental investigations are implemented to study the proposed RT system. Using Taguchi's method, the critical performance process parameters have been analyzed. The analysis of variance method is used to decide the important process parameters. The developed RT system can produce the mold inserts directly. After assembling the plenary injection mold, the Acrylonitrile butadiene styrene ABS plastic parts have been generated from this RT mold to demonstrate the success of the process. In the future, it is possible to develop a grading alloy RT system, which can control the ratio of the alloy grading layers. Because the grading alloy rapid mold can use varied metallic powders with varied ratio, the cooling time and the residual stress can be reduced. Furthermore, the RT system can also adopt the coaxial nozzle which can be integrated with the optical system. The advantage of the coaxial nozzle is the independence of the powder supply direction in which the workpiece moves

Rapid tooling (RT) has recently become more important due to the requirement of rapid manufacturing. RT can manufacture stronger mold and functional component than rapid prototyping (RP) does. Besides, functional component and production are valued gradually. Functionally graded materials (FGM) technology is proposed to apply to functional component. This study aims to propose a new path planning, contour path method, and control feed rate of two powder feeders with two DC server motors to design a functionally graded material (FGM) structure. The continuous material distribution in each layer is changed into discrete stepwise grading by contour path method. It can avoid the problem of manufacturing process with Zig-Zag and increase the resolution of the grading region. Furthermore, the powder delivery database of material is built to control the composition of FGM accurately. Finally, the results of path planning and the case of powder feed are successfully demonstrated

This paper presents a rapid statistic credence modeling methodology for ultrasonic range sensors based on a hybrid mechanical configuration. This configuration of ultrasonic range sensors will speed up a circular distance measurement around the robot within 180 milliseconds (resolution in 1deg). Besides, the sensor model provides the credence probability of actual points of reflection on obstacle target, and the convex or concave can be recognized by a simple vector transversal rule. This credence methodology with configuration could be incorporated as extra credence information on building grid map for environment exploration or provide the most reliable geometric beacons for self localization application and mapping, etc

Leveraging numerous integrated sensing devices placed close to the actual monitoring environments, the information that such networks can provide is more accurate and richer than the information provided with a system of few, expensive sensing devices. The general design of such sensor node is motionless, but the mobile sensor is useful in some aspects. Nodes can be distributed at any corner easily, move to a specific emergent area, or replace the power exhausted nodes. Mobile sensor nodes can increase the interaction between user and system. Moreover, the ability of auto-recharging for mobile sensor nodes can solve the powerlessness problem for sensor network. We had designed a mobile sensor node with compass to correct the direction of movement; six IR sensors let mobile nodes with the ability of obstacle avoidance. We propose an algorithm called triangular selection path planning (TSPP) method, and along with the dead-reckoning (DR) algorithm to merge heading readings from compass and two encoders mounted on the wheels to compute the dead-reckoned location estimation. With these technologies, our mobile sensor node can know their own position accurately and achieve the destination efficiently.

We have recently developed an intelligent security architecture in robot (ISR). Image system, security system, multi-interaction system and obstacle avoidance system can all communicate with each other through intelligent automation robot system (IA robot). We also discuss the modes and distance of obstacle avoidance that influence in the pathway of obstacle avoidance. In order to navigate IA robot to complete mission with obstacle avoidance system and security system by using IR sensors, and USB Web-camera installed in IA robot. Therefore, because the limitation of IR sensors and the action mode we have set, we have to choose a critical distance. When IA robot approaches an obstacle into this distance, ISR will start to avoid obstacle. We also provide seven kinds of functions installed on this IA robot to reach security service. We have successfully demonstrated the modes and distance of obstacle avoidance and the security system.

The location of nodes in sensor network is an important problem with application in resource allocation, location sensitive browsing, and emergency communications. A key problem in sensor network location is the creation of a method that is robust to measurement quantization and measurement noise and also has a reasonable implementation cost. The RSS and TDoA are the popular distance measurement methods, but can be easy affected by noise and not independent. This paper explores a covariance intersection (CI) that fuses together location estimations obtained from power propagation loss measurements and propagation time to obtain higher accuracy location estimate. The performances of Kalman filter type estimators are severely affected by the ignored cross covariance. CI algorithm provides a mechanism to fuse two or more random variables with unknown correlation such that the computed covariance of the new estimate is consistent; therefore CI method is quite suitable for sensor network application. In simulation result shows that position estimates are correct within small ranging distance with few initial master nodes of the system.

We design a "two-layered" dynamic power management strategy in our sensor node to reduce the power consumption. There are many factors such as the probability of event generation, the battery status, the coverage problem, and the communication situations in our consideration. In network communication, we propose an asynchronous awakening scheme so that each node is free to power on or off its components based on the observed event statistics, and make a trade-off between communication and power consumption. The deepest sleep state period is determined by taking account of the residual power. Combining with these methods, the lifetime of the sensor node can be extended as longer as possible without influencing the system performance.

In recent years, Rapid Tooling (RT) has become more important because of the requirement of rapid manufacturing. RT can be used to produce temporary mold, even permanent mold in mass production. RT makes more benefit from design stage to mass production. The objective of this research is to develop the Direct Metallic Rapid Tooling System and identify process parameters of laser cladding. A metallic powder feeder system, a three-axis motion control system, a 500W Nd-YAG laser system, a shielding gas supply system, and a powder nozzle are incorporated for the hardware architecture in Direct Metallic Rapid Tooling System. Besides, the designed software that includes slicing algorithm with bucket sorting, contour generation and tool path generation algorithm are presented briefly. Effort was attempted to find out the better process parameters of laser cladding using single clad and cladding layer techniques. Based on the investigation, some metallic RT parts were fabricated and discussed.

To keep autonomous mobile robot in continuously working condition, power supply is an important issue. Typically, rechargeable batteries may provide few hours of peak usage. Recharging is necessary before the power of the batteries has exhausted. In this paper, the docking station with an auto-recharging device and robot docking mechanism are implemented. We propose a docking control strategy and automatic recharging device for the security robot recharging. We also propose a power prediction algorithm to determine when the robot needs to navigate back to docking station for recharging. An artificial landmark is detected and recognized by the proposed image processing system. Then the geometrical relationship between the robot and the docking station (the depth and orientation) is estimated. The robot will move directly right in front of the docking station. Finally, the robot approaches to the docking station based on the proposed virtual spring model. In this model, we assume that the robot and the docking station are connected by a virtual spring. The compliant forces act both in the direction of the translation deformation and bending. The motion control parameters, which are velocities of the two wheels, can be derived from the model. Experimental results show that the proposed methods successfully perform the docking and automatic recharging.

Rapid tooling (RT) refers to the rapid creation of molds in much the same way as rapid prototyping, which means the rapid creation of models. Rapid prototyping is a method that offers both designers and manufacturers some advantages such as time compression and cost reduction. Rapid tooling has become more important because of the requirement of rapid manufacturing. Rapid tooling can be used to produce temporary mold, even permanent mold in mass production. Rapid tooling makes more benefit from design stage to mass production. The objective of This work is to develop a metallic rapid prototyping system and form a complete set of important parameters of laser cladding. The software technique includes slicing algorithm with bucket sorting and tool path generation algorithm. The hardware architecture includes 500 W Nd-YAG laser, metallic powder feeder, coaxial nozzle, shielding gas (Ar) supply system, CNC milling system and linear planar motor based working table. The system controller is a PC-based control system, which includes 3-axis AC servo motor driver and I/O control card. The system control software is written by Visual Basic language. The prototype system has been constructed and demonstrated.

A rapid prototyping (RP) machine system which combines a PC-based controller with the thermal-extrusion method is presented. The proposed RP system offers a three-axis platform, an extrusion head, a temperature controller, and a PC-based control system. Low-cost acrylonitrile-butadiene-styrene (ABS) pellets/powder are used for thermal extrusion, although the system is easily adapted to other not-too-dissimilar materials. In order to improve the quality of RP part, the Taguchi method was used to analyzing the process parameters of the proposed RP system. Based on the experimental results, the proposed RP mechatronics system can produce good quality RP parts. The RP software technique includes slicing, support, tool path, and motion code generation. This paper also presents a new adaptive slicing algorithm for RP system. According to this algorithm, the three-dimensional (3-D) computer-aided design (CAD) model can be sliced with different thickness automatically by comparing the contour circumference or the center of gravity of the contour with those of the adjacent layer. With this adaptive slicing method, the part can be fabricated faster than it uses uniform slicing method with identical accuracy. Finally, the intelligent web-based RP system allows remote users to upload an STL file, to directly building up of the physical model, and to monitor the actual fabrication process from a charged coupled devices (CCD) camera located in the RP machine itself. The user does not need to buy an expensive RP machine; instead, he can rent time and uses the machine remotely via the Internet.Note to Practitioners-RP has being used as an essential augmented tool between CAD and computer-aided manufacturing (CAM) for product manufacturing. RP uses layered manufacturing technology to produce complicated prototypes directly from a CAD model. As imaging, materials, and processing techniques improve, RP functions such as assembly fit, tooling masters, prototype tools, and prototype parts are made possible. RP technology that can construct parts of multiple materials, colors, and even parts composed of composite materials will soon be available. In this paper, an RP machine system which combines PC-based controller with thermal extrusion method is presented. Low-cost ABS pellets are used for thermal extrusion, although the system is easily adapted to other not-too-dissimilar materials. Based on the experimental results, the proposed RP mechatronics system can produce good quality RP parts. In this paper, a new adaptive slicing algorithm is developed to decrease fabrication time without much reducing the model accuracy. According to this algorithm, the 3-D CAD model can be sliced with different thickness automatically. With this adaptive slicing method, the part can be fabricated much faster than it uses traditional uniform slicing method. Finally, the intelligent web-based RP system has been developed which allows remote users to upload a CAD file of the part, direct building up of the physical model, and monitoring of the actual fabrication process from a CCD camera located in the RP machine itself. The greatest advantage is that user does not need to buy an expensive RP machine; instead he rents time and uses the machine remotely via the Internet. This really drastically reduces the production development cycle for the highly competitive time-to-market challenge.

The security of home, laboratory, office and factory is essential to human daily life. A danger event is often caused by the negligence of humans. Potential hazards may injure our life. Therefore, it motivates us to develop an intelligent multi-sensor based Security Robot system. It is expected to be widely employed in our daily life. Security robot can detect dangerous situation and provide timely alert us. The structure of the security robot contains eight parts. Including remote surveillance and control system, image system, obstacle avoidance system, software system, auto-dialing, driver system, sensor system and motion planning system. In this paper, we discuss the opportunity to use multi-agent technology in the sensory system and the expected improvements. The sensory system has seven-variety detection and diagnosis agent (local agent) and one sensor agent (auxiliary agent). Finally, we use multi-processor architecture to implement the multi-agent based sensory system for the security robot application.

Rapid prototyping is an automated manufacturing process that quickly builds physical models from CAD files of 3D prototypes. It dramatically speeds up design and manufacturing processes and substantially reduces the cost. With the highly advanced network technologies, many different software and hardware systems can be integrated as a distributed platform to provide more services, solve more complicated problems, and achieve better performance. We build a distributed Web platform on Internet using CORBA to effectively utilize various RP resources running on Linux, Solaris, or Microsoft Windows platforms. Users can design, simulate, and manufacture prototypes through a browser interface. Another objective of this paper is to develop a thermal extrusion based RP system which is equipped with a linear 2D planar motor. In order to improve the quality of RP part, Taguchi method was used to analyzing the process parameters of the proposed RP system. Based on the experimental results, the proposed RP mechatronics system can produce good quality RP parts. Based on the experimental results, the proposed RP system can produce good quality RP parts.

Intelligent robotic systems have been extensively applied in factory automation, space exploration, intelligent buildings, surgery, military service, and also in our daily life. Various remote control methods have been performed for intelligent robotic systems, such as radio, microwave, computer networks, etc. Nowadays, the computer network services have broadly used in our daily life, such as FTP, Telnet, the World Wide Web, e-mail, etc. Consequently, it is very convenient to use the Internet to control intelligent robot, and the users will increase in the future. In the past few years, many researchers have been using the Internet as a command transmission medium which can control the intelligent robot and obtain feedback signals. Although the Internet has many advantages in a variety of fields, using the Internet to control intelligent robots also has some limitations, such as the uncertain time-delay problem, the uncertain data-loss problem, and the data-transmission security problem. In the literature, many experts proposed various methods to solve these problems. This paper will discuss these methods and analyze the effects on the remote control systems caused by these problems. The intelligent robot can simultaneously present low-level navigational capabilities, medium-level self-positioning capabilities, high-level motion-planning capabilities, and the ability to be controlled through the Internet. The issues for controlling intelligent robots through the Internet will be discussed in terms of direct control, behavior programming control, supervisory control, and learning control. Finally, we enumerate some opportunities for the application of network-based intelligent robots, and present some successful examples of networked intelligent robots in our laboratory. Future trends and concluding remarks appear at the end of this paper.

Describes an algorithm and experimental results for detecting and isolating sensory failures for a fire detection system. In the fire system, we use two smoke sensors, two flame sensors and two temperature sensors to detect fire events. These sensors can be classified into two groups, and each group has one smoke sensor one flame sensor and one temperature sensor. The sensory failure and isolation techniques described in the paper are based on weight variation of the sensory adaptive fusion method. From the simulation and experimental implementation results, it demonstrates that the method can exactly find out which sensor is faulty and isolate it. That is to say, when a sensory failure occurs, the system can exactly locate the sensory failure.

Rapid prototyping (RP) has been widely known as being able to fabricate 3-D objects with complex geometric shapes. Rapid tooling has become more important because of the requirement of rapid manufacture. Rapid tooling can be used to produce a temporary mold, even a permanent mold in mass production. The objective of the paper is to develop a rapid tooling system. The software technique includes a slicing algorithm and tool path generation algorithm. The hardware architecture includes a 500 W Nd:YAG laser, metallic powder feeder, substrate preheat system, shielding gas (N/sub 2/) supply system, CNC milling system and linear planar motor based working table. The system controller is a PC-based control system which includes 4-axis DSP motion control and an I/O control card. The system control software is written in Visual Basic language. The experimental result shows that the linear 2-D planar motor based X-Y table has fast response and can make an accurate tool path.

Slicing process plays a very important role in layered manufacturing technology. A slicing algorithm decomposes the 3D CAD model into 2D layer contours that can be easily built. This paper presents a new adaptive slicing algorithm for a rapid prototyping system. According to this algorithm, the 3D CAD model can be sliced with different thickness automatically by comparing the contour circumference or the center of gravity of the contour with those of the adjacent layer. The part can be fabricated faster using this adaptive slicing method than using the uniform slicing method with identical accuracy. The proposed method was validated with computer simulation, and the part can be produced on the developed LCD panel display based rapid prototyping system.

The objective of the paper is to develop multilevel multisensor based decision fusion principles for an intelligent animal robot. The paper presents the design and implementation of a four-legged animal robot. The animal robot's head, body, and legs were constructed using aluminum material. The control/decision unit, driver unit and sensory unit were designed for the control of the animal robot. In addition, we use hierarchically organized control hardware and multilevel multisensor based fusion techniques to obtain a fused decision. Weight function and rule based algorithms are employed to fuse and integrate multisensor data. Preliminary experimental results indicate that the proposed methods can effectively fuse decisions for the animal robot.

Motion planning plays an important role in the field of autonomous mobile robots. We propose a hybrid intelligent system including the object-oriented knowledge base and adaptive motion planning (AMP) algorithm. There are four major procedures in the intelligent system: a priori environment building, on-line topology map generation, candidate path searching, and behavior commands generation. The adaptive motion planning (AMP) algorithm acquires the a priori defined map and the latest or dynamic information. The algorithm then determines path and generates a sequence of motion behavior description commands for navigation. The experimental results indicate that the object-oriented knowledge base can be updated easily. The AMP algorithm can produce the suitable path and behavior commands, and fuse these behaviors to navigate the mobile robot in the dynamic environment. The proposed method is implemented on the "Chung Cheng I" autonomous mobile robot to demonstrate the reliability and flexibility.

This paper presents a hierarchical grey-fuzzy motion decision-making (HGFMD) algorithm, which is capable of integrating multiple sequential data for decision making and for the design of the control kernel of the target tracking system. The algorithm combines multiple grey prediction modules, each of which can estimate a suitable model from sequential sensory information for approximating the observed dynamic system for future-trend prediction and for decision making through a multilayer fuzzy logic inference engine. We have designed the HGFMD controller for a target tracking system and implemented it in our autonomous mobile robot. The HGFMD is compared with the conventional fuzzy logic controller, multilayer fuzzy controller, and the original grey-fuzzy controller developed previously in various target-tracking experiments. We demonstrated the high reliability of the HGFMD controller and tracking system even when encountering the uncertain status of slow sensory response time and the nonlinear motion behaviors of the target.

This paper presents a hierarchical grey-fuzzy motion decision-making (HGFMD) algorithm, which is capable of integrating multiple sequential data for decision making and for the design of the control kernel of the target tracking system. The algorithm combines multiple grey prediction modules, each of them can online estimate the suitable model from sequential sensory information to approximate the observed dynamic system model for future-trend prediction, and for decision making through a multilayered fuzzy logic inference engine. We designed the HGFMD controller for target tracking system and implemented in our autonomous mobile robot. The HGFMD is compared with the conventional fuzzy logic controller, multilayered fuzzy controller, and the original grey-fuzzy controller we have developed previously in various target-tracking experiments. The results obtained show the high reliability of the HGFMD controller and tracking system.

Rapid prototyping has been widely known as being able to fabricate 3D object with complex geometric shapes. Rapid prototyping can shorten the product development cycle and improve the design process by providing rapid and effective feedback to the designer, thus help a company keep ahead in a rapidly changing market. This paper presents software and hardware technologies of building LCD display based rapid prototyping system. Software design includes slicing algorithm, LCD photomask display process, user interface and motion control program. Hardware configuration of this architecture includes LCD photomask, optical system, z-axis elevator, resin supply system and PC based control system. After slicing process, the cross section contour of each layer is transferred to the LCD projector that is used for photomask. The optical system emits parallel light upwardly through the mask to expose and solidify the entire layer at once. After one layer is completed, the platform with the part rises and a new layer of photopolymer is applied to the bottom plate. After the sweep roller wipes the resin, the LCD photomask displays the cross sectional contour of the next layer. At this moment, the platform moves downwards to the correct height, and the next layer is cured subsequently. This process is done repeatedly until the whole part is finished. This LCD display based rapid prototyping system can operate smoothly and quietly.

Presents an intelligent electrical walking rehabilitation system for rehabilitation automation. A set of force sensors to measure trainee's ground reaction force are arranged at the force plates. The gait phases are detected by a neural network which inputs the force measurements directly. To realize the system in practical applications, we represent the rehabilitation expert knowledge by fuzzy membership functions and rules. The velocity tuning commands to assist the rehabilitation process are generated through the fuzzy inference engine in each detection cycle. We have completed the integration of these techniques in experimental tests. The results demonstrate that the system is able to adapt to the trainee walking phase to increase the training effect.

Target tracking is useful in many robotics applications, such as assembly, autonomous navigation, etc. Conventional position based tracking strategies usually define the dynamics models for either motion control on tracker or position prediction on target, making the system inflexible and difficult to implement. This paper presents a new target tracking strategy, in which the dynamics model parameters are online updated according to the limited observation of the target and tracker relation. The grey theory is used in the position prediction of the target, which predicts the future trend based on the previous sensory measurements. We also use the look-ahead fuzzy logic for motion control of the tracker. Experimental results for tracking different moving targets are accomplished by the "Chung Cheng-1" mobile robot used in our laboratory. The results demonstrate the robustness and flexibility of the system.

Many potential accidents due to direct control of electrical wheelchair may be occurred by unreasonable control in various internal or external environmental conditions, such as the uncertainties of the front caster wheels and the unbalanced friction condition between wheels and floor. This paper presents a grey-fuzzy decision-making (GFD) algorithm based on grey prediction theory and fuzzy logic theory. The GFD is able to estimate optimal parameters approximating to the real system dynamics mode according to the observed information. The application of intelligent assistant control of an electrical wheelchair investigated. We conduct the experiments by multisensor based wheelchairs "Luoson#1" and "Luoson#3". The GFD controller is presented to integrate the electronic compass and dead reckoning measurements for the reduction of the influence caused by the uncertainties. The results have demonstrated the system capability for the increased certainty and safety to operate the electrical wheelchairs.

A rapid prototyping machine system which combines three axes PC-based controller with thermal extrusion method is presented. The main features of this system are to build up a physical object model through the extrusion of thin thermal plastic material and building it up layer by layer. User can use this system to turn CAD model into physical object model. Hardware design includes three axes working table, a thermal extrusion head, and a PC-based control system. The rapid prototyping system software includes CAD-model slicing process and 2D-layer cross sectional fill generation algorithms. The process begins with the design of a conceptual geometric CAD model of an object. The cross sectional STL format data is interfaced to the rapid prototyping machine system. The system can be extended to use other plastic materials from the thermal extrusion head and build up the model layer by layer.

Telecontrol of a rapid prototyping machine via the Internet for the application of a telemanufacturing system is presented. The main features of this system are that users just need a general-purpose computer and a World Wide Web browser for commanding the rapid prototyping machine in a remote location through the Internet and Home Page. Hardware configuration of this architecture includes a thermal extrusion based layer-deposition rapid prototyping machine and several controllers, each of them is able to communicate with others via a computer network. Software design includes the user interface, model slicing process, model support generation, 2D-layer fill generation and machine codes translation. The system has extensive capabilities for teleoperation and other multimedia services. The system allows users to directly control the rapid prototyping machine via the Internet and access our Home Page through a friendly user interface. Also the remote user can receive real time images e.g. the completed rapid prototyping model parts captured by a CCD camera that is mounted on the rapid prototyping machine.

The development of intelligent wheelchairs can bring a good solution to assist severely handicapped people who are unable to operate the classical electrical wheelchair by themselves in their daily activities. In this paper we have developed a new concept on automatic guided intelligent wheelchair system, through which the wheelchair users can extend their activity area from a priori known environment as many autonomous approaches addressed. We describe a hierarchical look-ahead grey-fuzzy decision motion algorithm for the tracking control of the wheelchair to follow a guiding service robot in an unknown environment. We have developed a multisensor based intelligent wheelchair prototype called Luoson III to work with the previous built autonomous guiding service robot Chung-Cheng 1. The experimental results have demonstrated the success for the motion decision algorithm and the whole concept.

A force reflective feedback control system for the operation of intelligent wheelchairs is presented. A multisensor based differential-driver intelligent wheelchair "Luoson III" has been developed to conduct the experiments. We have used a Microsoft Force Feedback Pro joystick which has the force feedback and programmable capability. The joystick not only can receive the use input but also can simulate the effects of wall, shake, constant force and reflect back to the user. We have demonstrated the success of this system through the prototype wheelchair operated in an open office environment.

A multiagent with event driven obstacle avoidance system (MEDOAS) has been developed, which is capable of avoiding stationary and moving obstacles and achieve the goal position. The multiagent design enables the mobile robot to perform sensing, steering and driving control in parallel. Event driven motion control of the system enables the robot to cope with various difficulties of the given assignment in a dynamic changing task environment. The MEDOAS is implemented by multi-thread programming on a time sharing operating system running on the main control system on our mobile robot system in the laboratory. The performance of MEDOAS is experimentally tested on the mobile robot. Experimental results demonstrate the resultant MEDOAS system capable of navigating a mobile robot through the cluttered environment efficiently.

The components for providing autonomous capabilities of a mobile robot are grouped into few basic modules, namely motion planner, motion executor, motion assistant, and behavior arbitrator. The primitive motion executors such as obstacle avoidance, goal following, wall following, docking, and path tracking for mobile robot navigation are developed in this paper. They are integrated with motion planner, motion assistants, and behavior arbitrator together based on decentralized control architecture with a hierarchical shared information memory. Event driven concept is used on switching the motion behaviors. The resultant intelligence for mobile robot navigation is capable of efficiently performing motion behavior and detecting environmental event in parallel to adapt dynamically changed environment. It also allows the human to program the motion behaviors in high level to complete a task. Experimental results on a real mobile robot have demonstrated the robustness of the motion executors and the overall system.

Obstacle avoidance and other functional behaviors for autonomous mobile robot navigation usually play an independent role in a robot system. The cooperation between two or more functional behavior modules is loose relatively. As the navigation system becomes more complicated, the codes for various functional behaviors may become immense and redundant. In this paper, we propose a multiagent with event driven obstacle avoidance system (MEDOAS) for mobile robot navigation. The MEDOAS decomposes a behavior module, such as obstacle avoidance, into several primitive processing agents. The agent cooperates with others by communicating through a shared information memory. A central controller switches the agent activity by event driving rules. Based on MEDOAS more behavior modules can be achieved by adding new primitive agent to cooperate with the existing agents. Experimental results demonstrate the MEDOAS system capable of navigating a mobile robot through the cluttered environment efficiently.

The objective of this paper is to describe the development of a distributed multiagent sensory information fusion system which integrates data from multiple sensors such as laser radar, weather radar, forward looking infrared sensors (FLIR) and associated important sensory information located in different agents to accomplish requested tasks cooperatively. We have developed a system based on the concept of offboard augmented theater surveillance (OBATS) and Internet network architecture to make the multiagent resource management system in practice. The architecture of distributed fusion system, the coordination strategy of the global decision agent (GDA) and the communication between service agents (SAs) and GDA have been developed. Furthermore, a scenario application has been implemented to demonstrate the concept of this multiagent system. Other potential applications have also been discussed.

A remote supervisory control architecture which combines computer network and an autonomous mobile robot is constructed. The main feature of this architecture is that users just need a general purpose computer and a World Wide Web browser, they can command the mobile robot in a remote location through Internet and Home Page. Hardware configuration of this architecture includes a mobile robot, as server workstation, and other computers, each of them is able to communicate with others via computer network. Software design includes assignment of the server, communication between the server and robot, and the user interface. The architecture has extensive capabilities for teleoperation, telerobotics operations, and other multimedia services. Based on this architecture we have implemented multilevel operations for the user on various remote applications. At present, the system allows users to operate our mobile robot via the Internet and a friendly user interface (via our Home Page) using supervisory and direct control modules and they can receive live images captured by a CCD camera which is mounted on the robot and other data from sensor measurement.

Approaches in global motion planning (GMP) and geometric beacon collection (for self-localization) using traversability vectors have been developed and implemented in both computer simulation and actual experiments on mobile robots. Both approaches are based on the same simple, modular, and multifunctional traversability vector (t-vector). Through implementation it has been found that t-vectors reduce the computational requirements to detect path obstructions, Euclidean optimal via-points, and geometric beacons, as well as to identify which features are visible to sensors. Environments can be static or dynamic and polygons are permitted to overlap (i.e., intersect or be nested). While the t-vector model does require that polygons be convex, it is a much simpler matter to decompose concave polygons into convex polygon sets than it is to require that polygons not overlap, which is required for many other GMP models. T-vectors also reduce the data size and complexity of standard V-graphs and variations thereof. This paper presents the t-vector model so that the reader can apply it to mobile robot GMP and self-localization.

Autonomous vehicles require that all processes be efficient in time, complexity and data storage. In fact, an ideal system employs multifunctional models where ever possible. This paper presents the region- and feature-based neural network (RFNN) as a viable pattern analysis process engine for solving a variety of problems with a single math model. The RFNN employs receptive fields and weight sharing which compensate for noise, minor phase shifts and occlusions. The RFNN also utilizes greedy adaptive learning rates and mature feature preservation to expedite the overall training process. A novel ad hoc approach called "shocking" is used to solve the instability problem inherent to greedy adaptive learning rates. The basic RFNN "feature" is grounded in computer vision morphology in that the neural network autonomously learns subpatterns unique to various problems. This paper comprehensively describes the flexible RFNN architecture and training process and presents two problems that can be solved by the RFNN: sensor pattern-recognition and traffic sign recognition.

Recovering depth and 3D position using known camera motion is useful for solving many problems in robotics such as object manipulation, navigation, and localization. Most existing techniques attempt to solve the general problem by recovering depth across the entire image. They oftentimes make linear dynamics assumptions or use poorly recoverable metrics such as optical flow. An estimation technique is presented that retains the nonlinear camera dynamics and provides an accurate 3D position estimate of selected targets within the environment. In practice, the technique is accurate enough to fully specify the placement of a gripper for object manipulation after forward (as opposed to lateral) movement of a single camera. The technique also provides probabilistic information regarding the estimate, which can be used to make intelligent, informed decisions. We employed this technique at the 1995 AAAI robot competition where our robot, Lola, placed 1st in Office Clean-up where it was able to pick up trash and recyclables and deposit them in the appropriate receptacles with great success. Results of our implementation on Lola are also presented.

An approach to global motion planning for autonomous mobile robots has been developed on the basis of traversability vectors (t-vectors). Through the overall course of this research it was found that t-vectors provide a utility, efficiency and mathematical stability for collision detection and visibility that cannot be matched by commonly used algebraic approaches in static and dynamic environments. This paper will show that t-vectors also impact global motion planning by identifying redundancies in visibility graphs (V-graphs) and expediting their construction. The result of eliminating redundant path segments is a streamlined version of the V-graph called the essential visibility graph (EVG). This paper will also show that the EVG offers a significant reduction in data storage requirements and complexity.

This paper first reviews the existing techniques of surface-plane intersection in computer-aided geometric design (CAGD). The vector fields are introduced for the detection of closed intersection loops. A distance function is defined from a surface patch to an intersecting plane and the gradient of the function defines a plane vector field. The Poincare index theorem is inconclusive in detecting critical points of plane vector field. The modified Poincare theorem is developed by extending the plane vector fields into three dimensions for the conclusive test of critical points. The initial points are obtained based on loop detection and subdivision. The resulting intersections are represented using biarc spline fitting scheme. The algorithm presented in the paper requires only the surface be parametrically defined and C/sup 2/ continuous.

An approach to global self-localization for autonomous mobile robots has been developed using self-organizing Kohonen neural networks. This approach categorizes discrete regions of space using mapped sonar data corrupted by noise of varied sources and ranges. Our approach is similar to optical character recognition (OCR) in that the mapped sonar data can, over time, assume the form of a character unique to that room. Hence, it is believed that an autonomous vehicle can be capable of determining which room it is in based on mapped sensory data ascertained by wandering through and exploring that room. With some pre-processing and a robust explore routine, the solution becomes time-, translation- and rotation-invariant.

A distributed heterogeneous network of fuzzy control agents has been developed for reactive behavior-based control of an autonomous mobile robot This methodology allows the authors' vehicle, MARGE, to perform realistic tasks in unstructured environments. Control actions for the robot are generated by a colony of independent agents that compete and cooperate to determine the emergent motion of the vehicle. Fuzzy rules are used to implement a diverse array of real-time functions within one simple development environment. The authors' multi-layer approach differs from other methods that perform the fuzzy inference mapping in one step. Real-time control without special hardware is possible by using a singleton centroid calculation and by breaking complex behaviors down into simple modules.>

This paper presents a method for cutter path generation by directly intersecting the part surface with a plane to generate cutter contact (CC) points. The surface-plane intersection is traced out using the marching method. The surface patch is subdivided based on a loop detection criterion, and all the initial points can be found along the boundaries of subpatches. The biarc curve fitting is employed to fit the largest sequential data points on the intersection curve. The cutter location (CL) data is obtained for profile machining by offsetting the CC biarc curve in surface normal direction. For a general free form surface, approximation is taken to generate plane biarc curves to represent the CL data points. The method is tested in a rapid prototyping system for production a of foot last.>

Outdoor mobile robots usually rely on visual landmarks for guidance and navigation purposes. However the task of locating these landmarks in cluttered environment is very complex. The perceived size of the landmarks is a function of the imaging distance of the robot. Also, the appearance of the landmark may vary due to changes in lighting, viewing direction and interfering objects. In this paper we present a fractal based approach to landmark detection. Fractals are inherently scale invariant; the model presented here is also invariant to changes in light intensity. We present the performance of the incremental fractional Brownian motion (IFBM) classifier for natural textures as well as a segmentation tool for natural scene images.>

A motion planning and self-referencing approach has been developed, simulated and applied to an actual robot. Although there are several novelties to these approaches, the fact that both are based on traversability vectors (t-vectors) is one aspect of this research that is unique. Through their application it has been found that t-vectors enhance the detection of path obstructions and geometric beacons and expedite the identification of features that are visible (or hearable) to sensors in both static and dynamic environments. T-vectors also reduce the data size and complexity of standard V-graphs and variations thereof. This paper provides the t-vector models step-by-step so that the reader will be able to apply them to mobile robot motion planning and self-referencing.>

When navigating through an environment, a mobile robot updates its position and orientation by searching for known objects called geometric beacons. This operation of comparing predicted sonar ranges with observed ones is referred to as self-referencing. There are two fundamental steps necessary for solving the self-referencing problem: how to represent the objects in the environment and how to search for landmarks based on this representation. This paper presents a method of mobile robot self-referencing where every object in the environment can be used as a landmark beacon. First, it shows how to interpret and predict sonar range readings through regional sampling and sensor window construction. Second, it justifies geometrically representing obstacles on the basis of surface information quality, compatibility with global motion planning, and the computer storage necessary to satisfy both criteria. Finally, it justifies the use of traversability vectors and configuration-space-time based on how they can be simultaneously used to plan motion globally and facilitate the search for geometric beacons.

The modeling and implementation of an innovative micro capacitive proximity sensor using micromachining technology are described. The proximity sensor works on the principle of fringing capacitance. The target object to be measured does not need to be part of the measuring system and can be either a conductor or a nonconductor. By selecting a set of geometrical parameters, this sensor is capable of precise measurement of proximity in the range from micrometers to millimeters. The capacitance change is quite significant for such small sensing structures, though it is only a fraction of a picofarad. The micro size of the sensor makes it possible to mount it in many space-constrained places.

A new approach using fractal based vision is presented to solve the problem of mobile robot navigation in outdoor environments. Mobile robots rely on landmarks such as mile markers and street signs for information on global position and local traffic conditions. Due to the motion of the robot, the location, size and orientation of the landmarks are varying. Also, other objects in the scene might partially occlude the landmark. Thus, a robust recognition system is required to recognize the landmarks that may be distorted by a combination of these effects. A new fractal model called incremental fractional Brownian motion (BM) model, is developed to locate these landmarks. A new neural network architecture, reconfigurable neural network (RNN), is developed to recognize the landmarks. The fractal model is shown to be invariant to changes in intensity of incident light. The landmark candidate regions detected by the ifBM model are analyzed by the RNN. New learning rules based on update normalization are developed to decrease learning time and increase system stability. The network also has the ability to learn new patterns with minimal retraining time. The network is tested with images of actual street signs that were distorted by scale changes, rotations and occlusions.

Stationary visual cameras, or what is termed vision, can be used to provide the information necessary for the control of multiple free-ranging auto- matic guided vehicles in order to expand the range of applications for which they are a high-performance and cost-effective transport alternative. The camera placement problem of global vision refers to determining, for a given facility layout, a minimal number of cameras necessary to implement global vision-based control and their positions in the facility. In this paper, the camera placement prob- lem of global vision is formulated and analyzed for its computational complexity and is shown to be NP-hard, and a convex hull-based greedy algorithm is described for its approximate solution. I. INTRODUCTION Stationary visual cameras, or what is termed vision, can be used to provide the information necessary for the control of free-ranging automatic guided vehicle (AGV) systems. Global vision provides a means to robustly control AGV systems composed of large numbers of low-cost free- ranging vehicles that are able to operate in unrestricted indoor environments. The sensory information used for the control of the free-ranging AGV system is provided by global vision and low-cost range sensors on board each vehicle. A system with these capabilities will expand the range of applications for which they are a high-performance and cost-effective material transport alternative. The development of a means to achieve these capabilities was the principal motivation for proposing global vision-based free-ranging AGV control (3). Global vision refers to the use of cameras placed at fixed locations in a workspace to extend the local sensing available on board each vehicle in a free-ranging AGV system. Information from the cameras is used to (a) monitor the workspace to detect and track potential obstacles both in the immediate vicinity of each AGV and over its intended path; (b) track each AGV along its intended path to bound errors in the vehicle's dead-reckoning sensors; (c) monitor the load aboard each AGV to detect positioning errors; and, as a side benefit, (d) provide video images of the entire workspace so that a human operator can monitor the status of operations throughout the facility. Figure 1 shows the major components of global vision- based AGV control. The cameras for global vision are placed at fixed locations throughout a facility. Their unprocessed video signals are sent both to TV screens, where they could be used to monitor the status of on-going operations, and to a central processing site, where global vision-based vehicle control takes place. The status of the AGV system is available as input to high-level transport control functions (e.g., vehicle dispatching). Although only two cameras are shown in the figure, more cameras would typically be used in practice. Each vehicle has a communications link with the central pro- cessing site to receive control commands and to transmit the results of its on-board sensing operations. A. Limitations of Current AGV Systems

In this paper we present a neural net- work based approach to landmark recognition. Landmarks are used by mobile robots for navi- gation and self referencing information. Due to the motion of the robot, the location and size of the landmark in the sensor is always changing. The network architecture presented here is capable of overcoming changes in scale as well as position of the landmark. The neu- ral network is directly fed with intensity im- ages. This is a significant reduction in compu- tation overhead since no high-level image pro- cessing is needed. Since robot motion in un- structured environments requires dynamic path planning, the detection algorithm must also be able to output guidance results. The network described here is able to output results in real time for the input patterns. We have trained the network to recognize common landmarks such as road signs (stop sign, yield sign, pedestrian-crossing sign, etc.). The algorithm has been successfully integrated into a mobile robot testbed developed at the Robotics and In- telligent Systems Laboratory at North Carolina State University.

A novel method for solving the motion planning problem in the presence of moving obstacles is presented. Compared to previous approaches, the proposed method is more general in that obstacles are allowed to move with translation and rotation at nonlinear velocity. Using path/velocity decomposition, the method emphasizes the development of an efficient algorithm for collision detection between the robot and moving polygons. This algorithm is devised using transversability vector theory to determine the geometric relationships between any two convex polygons or convex components of concave polygons. The interference between the robot and the moving obstacles is detected and registered on the constraint map to represent time constraints on the robot's motion along the path. The constraint map is then used for velocity planning using a simple algorithm. The problem of a robot crossing the street is used as an example for computational analysis.>

A parallel-jaw type multiple-d.o.f. robot hand is designed and implemented to manipulate workpieces within the hand by rotation and translation. An efficient manipulator control algorithm and part mating application is developed. A simulation result is also presented to show the changes of force during part insertion to verify the models developed.>

The objective of this study is to develop a method for mobile robot motion planning in the presence of moving obstacles. The concept of traversability vectors is used to analyze the spatial relationship between the robot and moving obstacles. Given a predefined path, the occupancy of the path by moving obstacles can be detected and registered on the constraint map. Obstacles on this map represent time constraints on the robot motion along the path. A search algorithm is then developed to coordinate the robot motion. Simulation results for this approach are discussed.>

An approach to representing and recognizing general 3D curved objects using range data is introduced. A 3D generalized Hough transformation which can be applied to general 3D curved-object recognition and which reduces the computation time and storage requirements is developed. The approach makes use of the relative geometric differences between particular points on the object surface and some virtual points which are prespecified arbitrarily and by task. The method has several advantages, such as a simple knowledge base with less storage requirement, flexibility of the database design according to the tasks to be performed, and the capability of compensating for the uncertainties of position estimation due to noise and quantization error. However, the algorithm may not be suitable for a purely planar object.>

The issues involved in integrating multiple sensors into the operation of a system are presented in the context of the type of information these sensors can uniquely provide. A survey is provided of the variety of approaches to the problem of multisensor integration and fusion that have appeared in the literature in recent years ranging from general paradigms, frameworks, and methods for integrating and fusing multisensory information to existing multisensor systems used in different areas of application. General multisensor fusion methods, sensor selection strategies, and world models are examined, along with approaches to the integration and fusion of information from combinations of different types of sensors. Short descriptions of the role of multisensor integration and fusion in the operation of a number of existing mobile robots are provided, together with proposed high-level multisensory representations suitable for mobile robot navigation and control. Existing multisensor systems for industrial and other applications are considered.>

The authors address the problem of robot multisensor fusion and integration with special emphasis on optimal estimation of fused sensor data. The investigation is based on a Unimation PUMA 560 robot and various external sensors. These include overhead vision, eye-in-hand vision, proximity, tactile array, position, force/torque, crossfire, overload, and slip sensing devices. The efficient fusion of data from different sources will enable the machine to respond promptly in dealing with the real world. Towards this goal, the general paradigm of a sensor data fusion system has been developed, and some simulation results as well as results from the actual implementation of certain concepts of sensor data fusion have been demonstrated.>

A robotic system that can visually track and intercept an arbitrary object which is traveling at an unknown velocity on a conveyor has been presented. A fiber-optic eye-in-hand vision system developed at NCSU is used as an integral part of the entire tracking system. The eye-in-hand system is used to characterize the object trajectory in real time, using a modified optical flow approach. A control strategy has been developed which utilizes the kinematic data that are extracted by the tracking algorithm to intercept the moving object. An overall system configuration and its basic principles are described. The demonstration of the initial results is presented.>

The objective of the authors is to develop an intelligent robot workstation capable of integrating data from multiple sensors. The investigation is based on a Unimation PUMA 560 robot and various external sensors. These include overhead vision, eye-in-hand vision, proximity, tactile array, position, force/torque, cross-fire, overload, and slip-sensing devices. The efficient fusion of data from different sources will enable the machine to respond promptly in dealing with the 'real world'. Towards this goal, the general paradigm of a sensor data fusion system has been developed, and some simulation results, as well as results from the actual implementation of certain concepts of sensor data fusion, have been demonstrated.>

In this paper, the usefulness of applying the complex moment, features for object identification using a tactile array sensor has been explored. Some complex moment invariants have been derived and implementation of those features has been conducted. With those moment invariants, we can eliminate the effect of lateral displacement and rotation from the tactile images. Through the generation of a decision tree and the utilization of the complex moment features, the shape of the objects from the tactile array sensor can easily be identified.

The objective of this paper is to develop an intelligent robot system through the integration of multiple sensors into robot tasks. The investigation is based on a Unimation PUMA 560 robot employing various external sensors. These sensors include overhead vision, eye-in-hand vision, proximity, tactile array, position, force/torque, cross-fire, overload and slip. The efficient machine representation of acquired sensory knowledge will enable robots to deal with the "real world". The general paradigm of a sensor data fusion system has been developed. This system will allows the robot to handle uncertainty in sensory data as well as to verify and recover sensor errors.