Mobile robots are increasingly deployed in shared environments where they must learn to navigate alongside humans. Deep Reinforcement Learning (DRL) techniques have shown promise in developing navigation policies that account for interactions within crowds, fostering socially acceptable movement. However, these techniques often depend heavily on collision avoidance rewards to ensure safe navigation. In this study, we introduce a novel reward component based on relative velocity for collision avoidance, which integrates both the robot's and humans' kinematics within personal distance constraints. We conducted a thorough evaluation comparing this new reward model against a conventional one in simulated environments using advanced DRL methods. Our findings indicate that the proposed reward model improves the robots' ability to avoid collisions and navigate towards their goals while being socially acceptable.

Microbial corrosion is considered the main reason for multi-billion dollar sewer asset degradation. Sewer pipe surface temperature is a vital parameter for predicting the micro-biologically induced concrete corrosion. Due to this important measure, a surface temperature sensor suite was recently developed and tested in an aggressive sewer environment. The sensors can fail and they may also put offline during the period of scheduled maintenance. In such situations, time series forecasting of sensor data can be an alternative measure for the operators managing the sewer network. In this regard, this paper focuses on the short-term forecasting of sensor measurements. The evaluation was carried out by forecasting the sensor measurements for different time periods and evaluated with different forecasting models. The ETS model leads to high short-term forecasting accuracy and the ARIMA model leads to high long-term forecasting accuracy. The models were evaluated on real data captured in a Sydney sewer.

Classification has become a vital task in modern machine learning and Artificial Intelligence applications, including smart sensing. Numerous machine learning techniques are available to perform classification. Similarly, numerous practices, such as feature selection (i.e., selection of a subset of descriptor variables that optimally describe the output), are available to improve classifier performance. In this paper, we consider the case of a given supervised learning classification task that has to be performed making use of continuous-valued features. It is assumed that an optimal subset of features has already been selected. Therefore, no further feature reduction, or feature addition, is to be carried out. Then, we attempt to improve the classification performance by passing the given feature set through a transformation that produces a new feature set which we have named the “Binary Spectrum”. Via a case study example done on some Pulsed Eddy Current sensor data captured from an infrastructure monitoring task, we demonstrate how the classification accuracy of a Support Vector Machine (SVM) classifier increases through the use of this Binary Spectrum feature, indicating the feature transformation's potential for broader usage.

Many standard robotic platforms are equipped with at least a fixed 2D laser range finder and a monocular camera. Although those platforms do not have sensors for 3D depth sensing capability, knowledge of depth is an essential part in many robotics activities. Therefore, recently, there is an increasing interest in depth estimation using monocular images. As this task is inherently ambiguous, the data-driven estimated depth might be unreliable in robotics applications. In this paper, we have attempted to improve the precision of monocular depth estimation by introducing 2D planar observation from the remaining laser range finder without extra cost. Specifically, we construct a dense reference map from the sparse laser range data, redefining the depth estimation task as estimating the distance between the real and the reference depth. To solve the problem, we construct a novel residual of residual neural network, and tightly combine the classification and regression losses for continuous depth estimation. Experimental results suggest that our method achieves considerable promotion compared to the state-of-the-art methods on both NYUD2 and KITTI, validating the effectiveness of our method on leveraging the additional sensory information. We further demonstrate the potential usage of our method in obstacle avoidance where our methodology provides comprehensive depth information compared to the solution using monocular camera or 2D laser range finder alone.

In order to achieve the ever increasing quantity and quality demands for agricultural products, technological innovations must be explored. In most cases, low cost and high quality sensing options are a top priority. This paper addresses the problem of predicting soil organic matter content in an agriculture field using information collected by a low-cost network of mobile, wireless and noisy sensors that can take discrete measurements in the environment. In this context, it is proposed that the spatial phenomenon of organic matter in soil to be monitored is modeled using Gaussian processes. The proposed model then enables the wireless sensor network to estimate the soil organic matter field at all unobserved locations of interest. The estimated values at predicted locations are highly comparable to those at corresponding points on a realistic image that is aerially taken by a very expensive and complex remote sensing system.

In recent years mobile robotic wireless sensor networks have been a popular choice for modelling spatial phenomena. This research is highly demanding and non-trivial due to challenges from both network and robotic aspects. In this paper, we address the spatial modelling of a physical phenomena with the network connectivity constraints while the mobile robots are striving to achieve the minimum modelling mismatch in terms of root mean square error (RMSE). We have resolved it through Gauss markov random field based approach which is a computationally efficient implementation of Gaussian processes. In this strategy, the Mobile Robotic Wireless Sensor Node (MRWSN) are centrally controlled to maintain the connectivity while minimizing the RMSE. Once the number of MRWSNs reach their maximum coverage, a new MRWSN is requested at the most informative location. The experimental results are convincing and they show the effectiveness of the algorithm.

Reliable sensing is a crucial factor for assessing the structural health conditions of civil infrastructure such as sewerage networks, which are susceptible to hydrogen sulfide induced concrete corrosion. In this context, this paper reports the work on the development and characterization of an instrumentation system using an infrared radiometer for monitoring surface temperature variations of the concrete through non-contact measurements. The surface temperature measurements are gathered by positioning the sensor at different distance and angles from the surface of interest. The effects of ambient lighting conditions during measurements are investigated. Furthermore, the sensing performance of the sensor is evaluated by performing statistical error analysis, and the efficacy of a custom-made signal processing board is tested by comparing the electrical signal with reference measures.

The main contribution of this paper is a high frequency, low-complexity, on-board visual-inertial odometry system for quadrotor micro air vehicles. The system consists of an extended Kalman filter (EKF) based state estimation algorithm that fuses information from a low cost MEMS inertial measurement unit acquired at 200Hz and VGA resolution images from a monocular camera at 50Hz. The dynamic model describing the quadrotor motion is employed in the estimation algorithm as a third source of information. Visual information is incorporated into the EKF by enforcing the epipolar constraint on features tracked between image pairs, avoiding the need to explicitly estimate the location of the tracked environmental features. Combined use of the dynamic model and epipolar constraints makes it possible to obtain drift free velocity and attitude estimates in the presence of both accelerometer and gyroscope biases. A strategy to deal with the unobservability that arises when the quadrotor is in hover is also provided. Experimental data from a real-time implementation of the system on a 50 gram embedded computer are presented in addition to the simulations to demonstrate the efficacy of the proposed system.

Scene classification is a fundamental perception task for environmental understanding in today's robotics. In this paper, we have attempted to exploit the use of popular machine learning technique of deep learning to enhance scene understanding, particularly in robotics applications. As scene images have larger diversity than the iconic object images, it is more challenging for deep learning methods to automatically learn features from scene images with less samples. Inspired by human scene understanding based on object knowledge, we address the problem of scene classification by encouraging deep neural networks to incorporate object-level information. This is implemented with a regularization of semantic segmentation. With only 5 thousand training images, as opposed to 2.5 million images, we show the proposed deep architecture achieves superior scene classification results to the state-of-the-art on a publicly available SUN RGB-D dataset. In addition, performance of semantic segmentation, the regularizer, also reaches a new record with refinement derived from predicted scene labels. Finally, we apply our model trained on SUN RGB-D dataset to a set of images captured in our university using a mobile robot, demonstrating the generalization ability of the proposed algorithm.

The performance of intelligent electromyogram (EMG)-driven prostheses, functioning as artificial alternatives to missing limbs, is influenced by several dynamic factors including: electrode position shift, varying muscle contraction level, forearm orientation, and limb position. The impact of these factors on EMG pattern recognition has been previously studied in isolation, with the combined effect of these factors being understudied. However, it is likely that a combination of these factors influences the accuracy. We investigated the combined effect of two dynamic factors, namely, forearm orientation and muscle contraction levels, on the generalizability of the EMG pattern recognition. A number of recent time- and frequency-domain EMG features were utilized to study the EMG classification accuracy. Twelve intact-limbed and one bilateral transradial (below-elbow) amputee subject were recruited. They performed six classes of wrist and hand movements at three muscular contraction levels with three forearm orientations (nine conditions). Results indicate that a classifier trained by features that quantify the angle, rather than amplitude, of the muscle activation patterns perform better than other feature sets across different contraction levels and forearm orientations. In addition, a classifier trained with the EMG signals collected at multiple forearm orientations with medium muscular contractions can generalize well and achieve classification accuracies of up to 91%. Furthermore, inclusion of an accelerometer to monitor wrist movement further improved the EMG classification accuracy. The results indicate that the proposed methodology has the potential to improve robustness of myoelectric pattern recognition.

This paper addresses the problem of selecting the most informative sensor locations out of all possible sensing positions in predicting spatial phenomena by using a wireless sensor network. The spatial field is modeled by Gaussian Markov random fields (GMRFs), where sparsity of the precision matrix enables the network to benefit from computation. A new spatial sensor selection criterion is proposed based on mutual information (MI) between random variables at selected locations and those at unselected locations and interested but unlikely sensor placed positions, which enhances resulting prediction. The GMRF-based optimality criterion is then proven to be computationally and efficiently resolved, especially in a large-scale sensor network, by a polynomial time approximation algorithm. More importantly, with demonstrations of monotonicity and submodularity properties of the MI set function in the proposed selection criterion, our near-optimal solution is also guaranteed by at least within(1-1/e) of the optimal performance. The effectiveness of the proposed approach is compared and illustrated using two real-life large data sets with promising results.

This paper presents a distributed spatial estimation and prediction approach to address the centrally-computed scheme of Gaussian Process regression at each robotic sensor in resource-constrained networks of mobile, wireless and noisy agents monitoring physical phenomena of interest. A mobile sensor independently estimate its own parameters using collective measurements from itself and local neighboring agents as they navigate through the environment. A spatially-distributed prediction algorithm is designed utilizing methods of Jacobi over-relaxation and discrete-time average consensus to enable a robotic sensor to update its estimation of obtaining the global model parameters and recursively compute the global goal of inference. A distributed navigation strategy is also considered to drive sensors to the most uncertain locations enhancing the quality of prediction and learning parameters. Experimental results in a real-world data set illustrate the effectiveness of the proposed approach and is highly comparable to those of the centralized scheme.

This paper proposes a data driven machine learning model for spatial prediction of hydrogen sulfide (H2S) in a gravity sewer system. The gaseous H2S in the overhead of the gravity sewer is modelled using a Gaussian Process with a new covariance function due to constraints of sewer boundaries. The covariance function is proposed based on the distance between two locations computed along the lengths of the sewer network. A mutual information based strategy is used to choose the best k sensor measurements and their locations from among n potential sensor observations and their locations. This provably NP-hard combinatorial sensor selection problem is addressed by maximizing the mutual information between the selected locations and the locations that are not selected or do not have any sensor deployments. A proof-of-concept study was carried out comparing the spatial prediction of H2S with a complex model currently used by Sydney Water. The proposed approach is shown to be effective in both modelling and predicting the H2S spatial concentrations in sewers as well as identifying optimal number of H2S sensors and their locations for a required level of prediction accuracy.

Human context is the most natural explanation why objects are placed and arranged in a particular order in an indoor environment. Usually, humans arrange objects in order to support their intended activities in a given environment. However, most of the common approaches for robotic object search involve modelling object-object relationships. In this paper, we hypothesize such relationships are centered around humans and bring human context to object search by modelling human-objects relationships through affordance-map. It identifies locations in a 3D map which support a particular affordance using virtual human models. Therefore, our approach does not require to observe real humans in the scene. The affordance-map and object-human-robot relationship are then used to infer the object search strategy. We tested our algorithm using a mobile robot that actively searched for the object "computer monitors" in an office environment with promising results.

This paper extends the recently developed Model-Aided Visual-Inertial Fusion (MA-VIF) technique for quadrotor Micro Air Vehicles (MAV) to deal with wind disturbances. The wind effects are explicitly modelled in the quadrotor dynamic equations excluding the unobservable wind velocity component. This is achieved by a nonlinear observability of the dynamic system with wind effects. We show that using the developed model, the vehicle pose and two components of the wind velocity vector can be simultaneously estimated with a monocular camera and an inertial measurement unit. We also show that the MA-VIF is reasonably tolerant to wind disturbances, even without explicit modelling of wind effects and explain the reasons for this behaviour. Experimental results using a Vicon motion capture system are presented to demonstrate the effectiveness of the proposed method and validate our claims.

This paper addresses the issue of monitoring physical spatial phenomena of interest utilizing the information collected by a network of mobile, wireless and noisy sensors that can take discrete measurements as they navigate through the environment. The spatial phenomenon is statistically modelled by a Gaussian Markov Random Field (GMRF) with hyperparameters that are learnt as the measurements accumulate over time. In this context, the GMRF approximately represents the spatial field on an irregular lattice of triangulation by exploiting a stochastic partial differential equation (SPDE) approach, which benefits remarkably in computation due to the sparsity of the precision matrix. A technique of the one-step-ahead forecast is employed to predict the future measurements that are required to find the optimal sampling locations. It is shown that optimizing the sampling path problem with the logarithm of the determinant either of a covariance matrix using a GP model or of a precision matrix using a GMRF model for mobile robotic wireless sensor networks (MRWSNs) even by a greedy algorithm is impractical. This paper proposes an efficient novel optimality criterion for the adaptive sampling strategy to find the most informative locations in taking future observations that minimize the uncertainty at unobserved locations. The computational complexity of our proposed method is linear, which makes the MRWSN scalable and practically feasible. The effectiveness of the proposed approach is compared and demonstrated using a pre-published data set with appealing results.

This paper presents a novel algorithm for fusing monocular vision and inertial information for quadrotor Micro Air Vehicles by incorporating the unique dynamic characteristics of that platform into the state estimation process. The dynamics of a quadrotor is unique in that a dual axis accelerometer mounted parallel to the propeller plane provides measurements that are directly proportional to vehicle velocities in that plane. By exploiting these dynamic characteristics, we show that all vehicle states, including the absolute scale, become observable in all motion patterns. This distinguishes our method with other visual-inertial fusion methods, which either assume zero accelerometer bias, or require “sufficiently exciting” motion, such as non-zero acceleration, to ensure observability of the scale. The advantages of our method over existing visual-inertial fusion algorithms are proved through a theoretical analysis using Lie Derivatives and verified by extensive simulations and experiments.

Simultaneous place classification and object detection (SPCOD) is an algorithm which is able to categorize the environment (place) and detect the objects presented in the environment. Although both place classification and object detection problems have been in discussion in literature, as a concept SPCOD is still in its early stage of research. Focusing mainly on the discrimination ability of SPCOD, in this paper we have proposed a pairwise conditional random field (CRF) framework to integrate mature techniques on laser data based place classification and vision based off-the-shelf object descriptor. Extensive experimental results on a public data set demonstrate the effectiveness of the proposed method.

Surface Electromyogram (EMG) signals are usually utilized as a control source for multifunction powered prostheses. A challenge that arises with the current demands of such prostheses is the ability to accurately control a large number of individual and combined fingers movements and to do so in a computationally efficient manner. As a response to such a challenge, we present a combined feature selection and projection algorithm, denoted as Mutual Components Analysis (MCA). The proposed MCA algorithm extends the well-known Principal Components Analysis (PCA) by pruning the noisy and redundant features before projecting the data. To implement the feature selection step, the mutual information concept is utilized to implement a new information gain evaluation function. The performance and significance of the proposed MCA is demonstrated on EMG datasets collected for the purpose of this research from eight subjects with eight electrodes attached on their forearm. Fifteen classes of fingers movements where considered in this paper with MCA achieving >95% accuracy on average across all subjects.

Understanding the environment in both geometric and semantic levels enables a robot to perform high-level tasks in complex environments. Therefore in recent years research towards identifying and semantically labeling the environments based on onboard sensors for mobile robots has been gaining popularity. After the era of heuristic and rule-based approaches, supervised learning algorithms like Support Vector Machines (SVM) and AdaBoost have been extensively used for this purpose showing satisfactory performance. With the introduction of graphical models, approaches like Conditional Random Fields (CRF) which take the advantage of connectivity of samples provide more flexibility to capture complex dependencies. In this paper, we focus on a real-world task which challenges the generalization ability of the model, evaluate some graph based features, propose a semi-supervised learning algorithm by iteratively utilizing the results from SVM and CRF, and suggest a solution for CRF parameter estimation with partially labeled training data. Experiments have been conducted on six real-world indoor environments demonstrating the competence of the algorithm.

The knowledge of terrain types is not only useful for improving passenger's safety and comfort, but also for bounding safe navigation routes for autonomous vehicles. Most of the work in the literature is limited to rover operations at slower speeds or rovers with rigid wheels. This paper discusses and presents a critical analysis of speed independency of classifiers for passenger vehicles with tyres at road operating speeds. The suitability of using different features calculated based on vertically mounted accelerometer data as well as road profile data were analyzed. The experimental results agree with the literature, where speed independency could be achieved by using the Fast Fourier Transform (FFT) features extracted from the estimated road profiles. However, we also show that it is possible to achieve comparable results by extracting different features directly from acceleration data. This eliminates the requirement of accurate knowledge of the vehicle model parameters or speed of the vehicle.

Discrete choice experiments have traditionally focused on improving the prediction of static choices that are measured through external reflection and surveys. It is argued that considering the underlying processes of decision making across a variety of contexts may further progress decision research. As a pilot study in this field, this paper explores the dynamic nature of decision-making by examining the associated brain activity, Electroencephalogram (EEG), of people while undertaking choices designed to elicit their preferences. To facilitate such a study, the Tobii-Studio eye tracker system was utilized to capture the participants' choice based preferences when they were observing seventy two sets of objects of three images offering potential personal computer backgrounds. Choice based preferences were identified by having the respondent click on their preferred image. In addition, the commercial Emotiv wireless EEG headset with 14 channels was utilized to capture the associated brain activity during the period of the experiments. Principal Component Analysis (PCA) was utilized to preprocess the EEG data before analyzing it with the Fast Fourier Transform (FFT) to observe the changes in the four principal frequency bands, theta (3 - 7 Hz), alpha (8 - 12 Hz), beta (13 - 30 Hz), and gamma (30 - 40 Hz). A mutual information (MI) measure was then used to study left-to-right hemisphere differences as well as front-to-back difference. Across six recruited participants there was a clear and significant change in the spectral activities taking place mainly in the frontal (theta and alpha across F3 and F4) and occipital (alpha and beta across O1 and O2) regions while the participants were indicating their preferences.

Representation of spaces including both geometric and semantic information enables a robot to perform high-level tasks in complex environments. Therefore, in recent years identifying and semantically labeling the environments based on onboard sensors has become an important competency for mobile robots. Supervised learning algorithms have been extensively used for this purpose with SVM-based solutions showing good generalization properties. The CRF-based approaches take the advantage of connectivity information of samples thereby provide a mechanism to capture complex dependencies. Blending the complementary strengths of Support Vector Machine (SVM) and Conditional Random Field (CRF), there have been algorithms to exploit the advantages of both to enhance the overall accuracy of place classification in indoor environments. However, experiments show that none of the above approaches deal well with diversified testing data. In this paper, we focus mainly on the generalization ability of the model and propose a semi-supervised learning strategy, which essentially improves the performance of the system. Experiments have been carried out on six real-world maps from different universities around the world and the results from rigorous testing demonstrate the feasibility of the approach.

One of the leading causes of automobile accidents is related to driving impairment due to drowsiness. A large percentage of these accidents occur due to drivers' unawareness of the degree of impairment. An automatic detection of drowsiness levels could lead to lower accidents and hence lower fatalities. However, the significant fluctuations of the drowsiness state within a short time poses a major challenge in this problem. In response to such a challenge, we present the Uncorrelated Fuzzy Locality Preserving Analysis (UFLPA) feature projection method. The proposed UFLPA utilizes the changes in driver behavior, by means of the corresponding Electroencephalogram (EEG), Electrooculogram (EOG), and Electrocardiogram (ECG) signals to extract a set of features that can highly discriminate between the different drowsiness levels. Unlike existing methods, the proposed UFLPA takes into consideration the fuzzy nature of the input measurements while preserving the local discriminant and manifold structures of the data. Additionally, UFLPA also utilizes Singular Value Decomposition (SVD) to avoid the singularity problem and produce a set of uncorrelated features. Experiments were performed on datasets collected from thirty-one subjects participating in a simulation driving test with practical results indicating the significance of the results achieved by UFLPA of 94%-95% accuracy on average across all subjects.

Robotic devices are increasingly penetrating the human work spaces as stand alone units and helpers. It is believed that a robot could be easily integrated with humans, if the robot can learn how to behave in a socially acceptable manner. This involves a robot to observe, learn and comply with basic rules of human behaviors. As an example, one would expect a robot to travel in an environment without intruding human workspaces unnecessarily. Thus, identifying common motion patterns of people by observing a specific environment is an important task as people's trajectories are usually not random, however are tailored to the way the environment is structured. We propose a learning algorithm to construct a Sampled Hidden Markov Model (SHMM) that captures behavior of people through observations and then demonstrate how this model could be exploited for planning socially aware paths. Experimental results are presented to demonstrate the viability of the proposed approach.

Human robot interaction is an emerging area of research, where human understandable robotic representations can play a major role. Knowledge of semantic labels of places can be used to effectively communicate with people and to develop efficient navigation solutions in complex environments. In this paper, we propose a new approach that enables a robot to learn and classify observations in an indoor environment using a labeled semantic grid map, which is similar to an Occupancy Grid like representation. Classification of the places based on data collected by laser range finder (LRF) is achieved through a machine learning approach, which implements logistic regression as a multi-class classifier. The classifier output is probabilistically fused using independent opinion pool strategy. Appealing experimental results are presented based on a data set gathered in various indoor scenarios.

This paper presents a statistically consistent SLAM algorithm where the environment is represented using a collection of B-Splines. The use of B-Splines allow environment to be represented without having to extract specific geometric features such as lines or points. Our previous work proposed a new observation model that enables raw measurements taken from a laser range finder to be transferred into relative position information between the control points of a B-Spline and the robot pose where the observation is made. One of the unresolved issues in the work was the estimation of the observation covariance, which is addressed through an analytical approach in this paper. As the uncertainty associated with the observation model is accurately defined and an optimization approach is used in the estimation process, the proposed SLAM algorithm can produce consistent estimates. Both simulation and experimental data are used for evaluation of the results.

Human robot interaction is an emerging area of research with many challenges. Knowledge about human behaviors could lead to more effective and efficient interactions of a robot in populated environments. This paper presents a probabilistic framework for the learning and representation of human motion patterns in an office environment. It is based on the observation that most human trajectories are not random. Instead people plan trajectories based on many considerations, such as social rules and path length. Motion patterns are learned using an incrementally growing Sampled Hidden Markov Model. This model has a number of interesting properties which can be of use in many applications. For example, the learned knowledge can be used to predict motion, infer social rules, thus improve a robot's operation and its interaction with people in a populated space. The proposed learning method is extensively validated in real world experiments.

Extending metric space representations of an environment with other high level information, such as semantic and topological representations enable a robotic device to efficiently operate in complex environments. This paper proposes a methodology for a robot to classify indoor environments into semantic categories. Classification task, using data collected from a laser range finder, is achieved by a machine learning approach based on the logistic regression algorithm. The classification is followed by a probabilistic temporal update of the semantic labels of places. The innovation here is that the new algorithm is able to classify parts of a single laser scan into different semantic labels rather than the conventional approach of gross categorization of locations based on the whole laser scan. We demonstrate the effectiveness of the algorithm using a data set available in the public domain.

Steering maneuver is essential in robotic motion planning. Despite a lot of steering mechanisms successfully developed in past years, for miniature robots, real-time computation is still a limitation for robot path tracking. The design issues in cooperative control of battery-powered nonholonomic robots rest with the complicacy of the control strategies, the low power consumption and real-time processing capability. Conventionally, the improvement of computing speed mostly relies on the increment of the system clock and often results in some transient loss. Thus, an elaborate control algorithm developed for PC might not work on an embedded system. This paper presents a comprehensive steering algorithm which, via issuing predicaments for computation, will dramatically reduce the resource usage in hardware circuit design. The proposed algorithm is implemented on an embedded system for ubiquitous robotics using the field programmable gate array (FPGA) technology.

Relating information originating from disparate sensors observing a given scene is a challenging task, particularly when an appropriate model of the environment or the behaviour of any particular object within it is not available. One possible strategy to address this task is to examine whether the sensor outputs contain information which can be attributed to a common cause. In this paper, we present an approach to localise this embedded common information through an indirect method of estimating mutual information between all signal sources. Ability of L1regularization to enforce sparseness of the solution is exploited to identify a subset of signals that are related to each other, from among a large number of sensor outputs. As opposed to the conventional L2regularization, the proposed method leads to faster convergence with much reduced spurious associations. Simulation and experimental results are presented to validate the findings.

The main contribution of this paper arises from the development of a new framework, which has its inspiration in the mechanics of human navigation, for solving the problem of Simultaneous Localization and Mapping (SLAM). The proposed framework has specific relevance to vision based SLAM, in particular, small baseline stereo vision based SLAM and addresses several key issues relevant to the particular sensor domain. Firstly, as observed in the authors' earlier work, the particular sensing device has a highly nonlinear observation model resulting in inconsistent state estimations when standard recursive estimators such as the Extended Kalman Filter (EKF) or the Unscented variants are used. Secondly, vision based approaches tend to have issues related to large feature density, narrow field of view and the potential requirement of maintaining large databases for vision based data association techniques. The proposed Multi Map SLAM solution addresses the filter inconsistency issue by formulating the SLAM problem as a nonlinear batch optimization. Feature management is addressed through a two tier map representation. The two maps have unique attributes assigned to them. The Global Map (GM) is a compact global representation of the robots environment and the Local Map (LM) is exclusively used for low-level navigation between local points in the robot's navigation horizon.

Providing safety critical information to the driver is vital in reducing road accidents, especially at intersections. Intersections are complex to deal with due to the presence of large number of vehicle and pedestrian activities, and possible occlusions. Information available from only the sensors onboard a vehicle has limited value in this scenario. In this paper, we propose to utilize sensors on-board the vehicle of interest as well as the sensors that are mounted on nearby vehicles to enhance the driver situation awareness. The resulting major research challenge of sensor registration with moving observers is solved using a mutual information based technique. The response of the sensors to common causes are identified and exploited for computing their unknown relative locations. Experimental results, for a mock up traffic intersection in which mobile robots equipped with laser range finders are used, are presented to demonstrate the efficacy of the proposed technique.

Most of the lane marking detection algorithms reported in the literature are suitable for highway scenarios. This paper presents a novel clustered particle filter based approach to lane detection, which is suitable for urban streets in normal traffic conditions. Furthermore, a quality measure for the detection is calculated as a measure of reliability. The core of this approach is the usage of weak models, i.e. the avoidance of strong assumptions about the road geometry. Experiments were carried out in Sydney urban areas with a vehicle mounted laser range scanner and a ccd camera. Through experimentations, we have shown that a clustered particle filter can be used to efficiently extract lane markings.

In the past few years, there has been significant advancement in localization and mapping using stereo cameras. Despite the recent successes, reliably generating an accurate geometric map of a large indoor area using stereo vision still poses significant challenges due to the accuracy and reliability of depth information especially with small baselines. Most stereo vision based applications presented to date have used medium to large baseline stereo cameras with Gaussian error models. Here we make an attempt to analyze the significance of errors in small baseline (usually <0.1m) stereo cameras and the validity of the Gaussian assumption used in the implementation of Kalman filter based SLAM algorithms. Sensor errors are analyzed through experimentations carried out in the form of a robotic mapping. Then we show that SLAM solutions based on the extended Kalman filter (EKF) could become inconsistent due to the nature of the observation models used

Multimodal sensor registration and calibration are crucially important aspects in distributed sensor fusion. Unknown relationships of sensors and joint probability distribution between sensory signals make the sensor fusion nontrivial. In this paper, we adopt a Mutual Information (MI) based approach for sensor registration and calibration. It is based on unsupervised learning of a nonparametric sensing model by maximizing mutual information between signal streams. Experiments were carried out in an office like environment with two laser sensors capturing arbitrarily moving people. Attributes of the moving targets are used. Problems due to target occlusions are alleviated by the multiple model tracker. The registration and calibration methodology does not require any artificially generated patterns or motions unlike other calibration methodologies

Knowledge of calibration, that defines the location of sensors relative to each other, and registration, that relates sensor response due to the same physical phenomena, are essential in order to be able to fuse information from multiple sensors. In this paper, a mutual information (MI) based approach for automatic sensor registration and calibration is presented. Unsupervised learning of a nonparametric sensing model by maximizing mutual information between signal streams is used to relate information from different sensors, allowing unknown sensor registration and calibration to be determined. Experiments conducted in an office environment are used to illustrate the effectiveness of the proposed technique. Two laser sensors are used to capture people mobbing in an arbitrarily manner in the environment and MI from a number of attributes of the motion are used for relating the signal streams from the sensors. Thus the sensor registration and calibration is achieved without using artificial patterns or pre-specified motions

With limited dynamic range and poor noise performance, cameras still pose considerable challenges in the application of range sensors in the context of robotic navigation, especially in the implementation of simultaneous localisation and mapping (SLAM) with sparse features. This paper presents a combination of methods in solving the SLAM problem in a constricted indoor environment using small baseline stereo vision. Main contributions include a feature selection and tracking algorithm, a stereo noise filter, a robust feature validation algorithm and a multiple hypotheses adaptive window positioning method in 'closing the loop'. These methods take a novel approach in that information from the image processing and robotic navigation domains are used in tandem to augment each other. Experimental results including a real-time implementation in an office-like environment are also presented

Road-boundary detection is an integral and important function in advanced driver-assistance systems and autonomous vehicle navigation systems. A prominent feature of roads in urban, semi-urban, and similar environments, such as in theme parks, campus sites, industrial estates, science parks, and the like, is curbs on either side defining the road's boundary. Although vision is the most common and popular sensing modality used by researchers and automotive manufacturers for road-lane detection, it can pose formidable challenges in detecting road curbs under poor illumination, bad weather, and complex driving environments. This paper proposes a novel method based on extended Kalman filtering for fast detection and tracking of road curbs using successive range/bearing readings obtained from a scanning two-dimensional ladar measurement system. As compared with millimeter wave radar methods reported in the literature, the proposed technique is simpler and computationally more efficient. This is the first of its kind reported in the literature. Qualitative experimental results are presented from the application of the technique to a campus site environment to demonstrate the viability, effectiveness, and robustness.

In most urban roads, and similar environments such as in the theme parks, campus sites, industrial estates, science parks and the like, the painted lane markings that exist may not be easily discernible by CCD cameras due to poor lighting, bad weather conditions and inadequate maintenance. An important feature of roads in such environments is the existence of pavements or curbs on either side defining the road boundaries. These curbs, which are mostly parallel to the road, can be harnessed to extract useful features of the road for implementing autonomous navigation or driver assistance systems. However, extraction of the curb or road edge feature using vision image data is a difficult task as curbs are not conspicuous in the vision image. To extract the curb from a camera image requires extensive image processing, heuristics and very favorable lighting. In our approach, road curbs are extracted speedily using range data provided by a 2D laser measurement system (LMS). Experimental results are presented to demonstrate the viability, and effectiveness, of the proposed methodology and its robustness to different obstacle, weather and lighting conditions.

In most urban roads, and similar environments such as in theme parks, campus sites, industrial estates, science parks and the like, the painted lane markings that exist may not be easily discernible by CCD cameras due to poor lighting, bad weather conditions, and inadequate maintenance. An important feature of roads in such environments is the existence of pavements or curbs on either side defining the road boundaries. These curbs, which are mostly parallel to the road, can be harnessed to extract useful features of the road for implementing autonomous navigation or driver assistance systems. However, extraction of the curb or road edge feature using vision image data is a very formidable task as the curb is not conspicuous in the vision image. To extract the curb using vision data requires extensive image processing, heuristics and very favourable ambient lighting. In our approach, the curb data is extracted speedily using range data provided by a 2D laser range measurement device. This information is then used to extract the mid-line(s) in the vision image using an extended Kalman filtering (EKF) approach. Subsequently midline data is used for the prediction of the road boundaries. Experimental results are presented to demonstrate the viability, and effectiveness, of the proposed methodology.

In most urban roads and similar environments, such as in theme parks, campus sites, industrial estates, science parks and the like, the painted lane markings that exist may not be easily discernible by CCD cameras due to poor lighting, bad weather conditions, and inadequate maintenance. An important feature of roads in such environments is the existence of pavements or curbs on either side defining the road boundaries. These curbs, which are mostly parallel to the road, can be harnessed to extract useful features of the road for implementing autonomous navigation or driver assistance systems. However, extraction of the curb or road edge feature using vision image data is a very formidable task as the curb is not conspicuous in the vision image. To extract the curb using vision data requires extensive image processing, heuristics and very favorable ambient lighting. In our approach, road curbs are extracted speedily using the range data provided by a 2D laser range measurement system (LMS). Experimental results are presented to demonstrate the viability and effectiveness of the proposed methodology and its robustness to different road configurations including road intersections.

This paper presents a methodology for extracting road edge and lane information for smart and intelligent navigation of vehicles. The range information provided by a fast laser range-measuring device is processed by an extended Kalman filter to extract the road edge or curb information. The resultant road edge information is used to aid in the extraction of the lane boundary from a CCD camera image. The Hough transform is used to extract the candidate of lane boundary edges, and the most probable lane boundary is determined by using an active line model and minimizing an appropriate energy function. Experimental results are presented to demonstrate the effectiveness of the combined laser and vision strategy for road-edge and lane boundary detection.

Fuzzy logic control is a relatively new technology and hence it needs rigorous comparative analyses with other well-established conventional control schemes. Furthermore, fuzzy controller stability analysis is a major hindrance for its popularity among control engineers. This paper shows how stable fuzzy controllers may be synthesized for a typical AGV from the perspective of variable structure systems (VSS) theory. VSS or sliding model control (SMC) is an established robust nonlinear control methodology. The AGV is characterized by highly nonlinear, coupled and configuration dependent dynamics, with uncertainty in model parameters. Similarity in performance of the fuzzy controllers to the SMC controller is demonstrated through experimental results obtained for steering control of the AGV.