Algorithmic bias often arises as a result of differential subgroup validity, in which predictive relationships vary across groups. For example, in toxic language detection, comments targeting different demographic groups can vary markedly across groups. In such settings, trained models can be dominated by the relationships that best fit the majority group, leading to disparate performance. We propose framing toxicity detection as multi-task learning (MTL), allowing a model to specialize on the relationships that are relevant to each demographic group while also leveraging shared properties across groups. With toxicity detection, each task corresponds to identifying toxicity against a particular demographic group. However, traditional MTL requires labels for all tasks to be present for every data point. To address this, we propose Conditional MTL (CondMTL), wherein only training examples relevant to the given demographic group are considered by the loss function. This lets us learn group specific representations in each branch which are not cross contaminated by irrelevant labels. Results on synthetic and real data show that using CondMTL improves predictive recall over various baselines in general and for the minority demographic group in particular, while having similar overall accuracy.

A recent proliferation of mobile devices led to immense deployment of WLAN access points (APs). IEEE 802.11 mobile devices periodically perform active background scanning to discover available APs in the vicinity. However, this periodic active scanning imposes unnecessary overhead in a static WLAN environment. This paper addresses this scanning overhead issue of mobile devices in IEEE 802.11 WLANs and proposes a smart scanning method using a motion sensor. Our scheme scans channels only if the signal strength from the associated AP becomes below a specified threshold and a mobile device is moving. The smart scanning suppresses unnecessary periodic background scans. We implemented the smart scanning in a commercial smartphone and evaluated its performance in a real indoor WLAN environment. The experimental and simulation results show that the smart scanning can significantly reduce the scanning overhead, the number of unnecessary handoffs and force disconnections.

Although the latest video codecs such as H.264/AVC, H.264/SVC, and VP8 were developed with network-friendly features, provisioning the quality of video delivery over IEEE 802.11 wireless networks is still challenging because of error-prone medium and random access. This paper evaluates the quality of video delivery using those latest video codecs over IEEE 802.11. Our results show that reduction of coded video data, consideration of the queue size of each Access Category, and the mandatory implementation of error recovery features in H.264/SVC can help provide robust multimedia transmission over IEEE 802.11. In addition, the performance of the latest video codecs is compared via extensive simulation scenarios. Our comparison shows that VP8 achieves good video quality with the basic medium access control techniques of IEEE 802.11, whereas H.264/AVC and H.264/SVC benefit significantly from appropriate mapping schemes between encoded video fragments and IEEE 802.11e Access Categories.

Providing seamless multimedia services to mobile users is one of the important requirements for the current Internet evolution. However, due to the effects of host mobility, it is difficult to achieve seamless QoS guarantees in mobile Internet. Though Next Step In Signaling (NSIS) protocol was recently proposed as an alternative of RSVP, its applicability has not been completely verified particularly in mobile environment. This paper enhances NSIS protocol based on advance resource reservation using layers 2 and 3 complementation for supporting host mobility. Our approach detects a crossover node (CRN) and reserves network resources proactively along the new path which will be used after handoff This significantly reduces the latency of signaling session re-establishment caused by handoff. Only a few modifications are required to the current NSIS protocol for our approach. The experimental results demonstrate that our approach can provide seamless QoS guarantees for multimedia services in mobile Internet.

This paper presents mobile robot localization using coded infrared light as artificial landmarks. Different from RFID, identification using infrared light has highly deterministic characteristics. IRID(infrared identification) is implemented with IR LEDs and photo transistors. By putting several infrared LEDs on the ceiling, the floor is divided into several sectors and each sector is set to have a unique identification. The coded infrared light tells which sector the robot is in, but the size of the uncertainty is still too large if the sector size is large, which usually occur. Dead-reckoning provides the estimated robot configuration but the error is getting accumulated as the robot travels. This paper presents an algorithm which fuses both the encoder and the IRID information so that the size of the uncertainty becomes smaller. It also introduces a framework which can be used with other types of the artificial landmarks. The characteristics of the developed IRID and the proposed algorithm are verified from the experiments.

This paper describes a new sensor system for improving the accuracy of the range information using multiple IR range sensors. Environment and obstacle sensing is the key issue for mobile robot localization and navigation. Laser scanners cover 180deg and are accurate but too expensive. Radial range sensors such as laser scanners, IR scanners and ultrasonic range sensor rings have blind spots so that a small obstacle not close enough to the sensor may be easily missed. It is necessary to develop a low cost sensor system which covers 360deg with small blind spots. A sensor system with 12 IR range sensors is designed and implemented. The problem of the 2D sensor systems is, they measure the nearby obstacles on a plane where the sensors are located. In order to scan the environment vertically, a 2.5D sensor system is designed and the experimental result shows the effectiveness of the proposed sensor system

Two human-robot interactions, including a haptic interaction and a teleoperated interaction, are explored with a new exoskeleton-type masterarm, in which the electric brakes with the torque sensor beams are used for force reflection. In the haptic interaction with virtual environment, the masterarm is used as a haptic device and tested to examine how the resistant torque of the electric brake for the force reflection is implemented in contact regime prior to conducting the teleoperated interaction. Two types of virtual environments, a rigid wall with high stiffness (hard contact with 10 [KN/m]) and a soft wall with low stiffness (soft contact with 0.1 [N/m]), are integrated with the masterarm for the haptic interaction. In hard contact, large force is fed back to the human operator, and makes the human operator hardly move. The electric brake with the torque sensor beam can detect the torque and its direction so that it allows free motion as well as contact motion by releasing or holding the movement of the operator. The experimental results show how the electric brake is switched from contact to free regime to allow the operator to move freely, especially when the operator intends to move toward the free regime in contact. In soft contact, the force applied to the human operator can be increased or decreased proportionally to the torque amount sensed by the torque sensor beam, thus the operator can feel the contact force proportional to the amount of the deformation during the contact. Finally, the masterarm is integrated with the humanoid robot, CENTAUR developed at Korea Institute of Science and Technology to conduct a pick-and-place task through the teleoperated interaction. It is examined that the CENTAUR as a slave robot can follow the movement of the operator.

Open-loop estimation methods are commonly used in mobile robot applications. Their strength lies in the speed and simplicity of an estimate. However, these methods can sometimes lead to inaccurate or unreliable positional estimates. Using one or more optical flow sensors, a method has been developed which can accurately track position in both ideal kinematic conditions and otherwise. Using optical flow techniques and available sensors, reliable positional estimates are made. The sensor provides accurate measurement of the movement at the sensor location. Even though the sensor does not provide angular displacement, the robot movement is estimated using only one sensor even with wheel slip. However, when the robot moves sideways due to external disturbance, redundant sensors are used in order to estimate the configuration of the robot. Pseudo inverse based estimation and the extended Kalman filter based estimation are presented to show the effectiveness of the proposed approach. Location of the sensors has also been investigated in order to minimize errors caused by inaccurate sensor readings. Finally, the method is implemented and tested using a potential field based navigation scheme.

An accurate and compact map is essential to an autonomous mobile robot system. A topological map represents the environment in terms of the discrete nodes with edges connecting them. It is usually constructed by the Voronoi-like graphs. In this paper the topological map is incrementally built based on the local grid map using the thinning algorithm. This thinning-based topological map does not create the boundary edges and weak meet points which are found in the generalized Voronoi graph. Furthermore, the map can be built in real-time and is robust to the environment change. Since lack of metric data in the topological map poses difficulty in localization, the metric features such as corners are incorporated into the topological map, thus leading to the hybrid map. In this paper the detailed procedure to obtain this hybrid map is discussed and the experimental results are shown to verify the validity of the proposed algorithm

This paper is a summary of an effort to develop a powerful motion planning algorithm for the mobile manipulator. The mobile manipulator is expected to work in partially defined or unstructured environments. In our global/local approach to path planning, joint trajectories are generated for a desired Cartesian space path, designed by the global path planner. For local path planner, inverse kinematics for a redundant system is used. Obstacle avoidance and joint displacement limits for the manipulator links are considered in the motion planner. In an event of failure to obtain feasible trajectories, the task can not be accomplished. In the case of the joint constraint violation, use of Jacobian matrix element as gradient is proposed. At the point of failure, a derivation in the Cartesian space path is obtained and the re-planner gives a new path that would achieve the goal position. To calculate the deviation, a non-linear optimization problem is formulated and solved by standard sequential quadratic programming (SQP) method.

This paper presents the obstacle detection algorithm based on the consecutive range sensor scanning scheme and the probabilistic navigation for an mobile robot. For a known environment, a mobile robot scans the environment using the range sensor which can rotate 360/spl deg/. The environment is rebuilt using nodes of two adjacent walls, and an obstacle is detected by comparing characteristic points of both the known environment and the scanned data set. It is very useful for detecting the moving obstacle. By comparing two data sets, the movement of an obstacle is extracted. Furthermore, the consecutive scanning data set provides obstacle information in unknown environment. Geometric comparison between the two consecutive data sets is used to detect the obstacles and the algorithm is presented with both simulation and experimental results. After the obstacle information is extracted from the consecutive scanning, the path is rebuilt by checking the collision probability.

An approach for multiple mobile robot localization using inexpensive infrared sensors and CMOS cameras is presented. A multiple robot system is treated as a virtual linked robot. Using the virtual link length and the relative heading information in conjunction with sensor scan data for each robot in the system, the entire system can effectively be localized without investing heavily in sensors. In order to facilitate collaboration and exploit geometry, a team must be able to extract salient information from the collective sensing of all its members. By collecting and fusing sensor information from multiple, distinct positions, the effective resolution of the space can be improved beyond that of the individual measurements. The proposed localization method is verified via both simulation and experiment.

Open-loop estimation methods are commonly used in mobile robot applications. Their strength lies in the speed and simplicity of an estimate. However, these methods can sometimes lead to inaccurate or unreliable positional estimates. Using one or more optical flow sensors, a method has been developed which can accurately track position in both ideal kinematic conditions and otherwise. Using optical flow techniques and available sensors, reliable positional estimates are made. Location of the sensors has also been investigated in order to minimize errors caused by inaccurate sensor readings. Finally, the method is implemented and tested using a potential filed based navigation scheme. Estimates of position were found to be as accurate as dead-reckoning in ideal conditions and much more accurate in cases where kinematic violations (such as wheel slip) occurred.

We propose a novel technique for acquiring effective information for obstacle avoidance in mobile robot navigation. Instead of simply receiving data at a single point, we actively give perturbations to the system and measure the response from the system. Correlating the input and the output, we can formulate the correlation function and useful information to guide the mobile robot such as gradient can be obtained from this function. This algorithm is applied to the obstacle avoidance in mobile robot navigation and the experimental results are provided.

We describe a method for navigation in a known indoor environment, such as a home or office, that requires only inexpensive range sensors. Our framework includes a high-level planner which integrates and coordinates path planning and localization modules with the aid of a module for computing regions which are expected, with high probability, to contain the robot at any given time. The localization method is based on simple geometric properties of the environment which are computed during a preprocessing stage. The roadmap-based path planner enables one to select routes, and subgoals along those routes, that will facilitate localization and other optimization criteria. In addition, our framework enables one to quickly plan new routes, dynamically, based on the current position as computed by intermediate localization operations. We present simulation and hardware experimental results that illustrate the practicality and potential of our approach.

A new concept of the exoskeleton-type masterarm, composed of serial links, is introduced in the paper. To provide maximum range of human motion, several redundant joints are added to the serial links. In order to reduce the number of joints to be measured, kinematics of serial links was taken into consideration in design. Three measurable, controllable joints and three redundant free joints are used for the upper arm (shoulder), similarly to the forearm (wrist) while one measurable, controllable joint is used for the elbow. In particular, a torque sensor beam is designed for fine force reflection using the strain gauge. It detects the torque as well as its direction applied by the human operator, which allows the electric brake to be used as an actuator for force reflection. The electric brake constrains the joint movement so that the operator can feel the force. This electric brake outperforms the motor in terms of torque/weight ratio and makes the device light and compact. This masterarm measures the movement of the operator's arm precisely, and its can be used for teleoperation with a slave robot, or as a motion planner for an industrial robot.

Presents a rapid localization method for mobile robots. Localization, i.e., absolute position measurement, is an important issue since odometer errors render it impossible for any robot to precisely follow a specified trajectory, resulting in a growing difference between the actual configuration and the calculated configuration as the robot travels. Periodic localization is required to correct these errors. We propose a localization method using range sensor data which is based on simple geometric properties of the environment. In many common situations, information regarding the environment is provided a priori for path planning. During processing, the method proposed here utilizes this information to partition the workspace into sectors using simple visibility computations, and a small identifying label is computed for each sector. The localizer analyzes range sensor readings (distances) and extracts characteristic points, which are compared with the pre-computed sector labels to localize the robot, first to a sector, and then to a particular configuration within that sector. Advantages of this two step process are that it is computationally very simple, and that it allows precise localization without any landmarks from any configuration in the environment. This localization method also provides opportunities for the global navigation procedure to analyze and select trajectories in terms of their tolerance to localization errors.

For robot teleoperation, many researches have attempted to control the slave robot from remote site using a master arm. A new concept of the distributed controller architecture for the master arm is introduced. To reduce possible non-uniform time delay and to have higher position command update rate, the host controller with the distributed satellite controllers are developed. Each satellite controller measures the corresponding joint angle, while the host controller performs initial calibration, inverse kinematics calculation, and position commands generation. The host controller and the satellite controllers are networked via the SPI (serial peripheral interface) protocol, with maximum speed of 5Mbps. For force feedback, the host controller solves the joint torque relations based on kinematics of the master arm, and the satellite controller controls each of the joint torque. This distributed controller greatly reduces calculation and control loads and makes the electrical wiring very simple.

For robot teleoperation, a lot of research has been done to control the slave robot from a remote site using a masterarm. The concept of a distributed controller architecture for the masterarm is introduced in the paper. To reduce possible non-uniform time delay and to have higher position command update rate, the host controller with distributed satellite controllers is developed. Each satellite controller measures the corresponding joint angle, while the host controller performs initial calibration, inverse kinematics calculation, and position commands generation. The host controller and the satellite controllers are networked via a SPI (serial peripheral interface) protocol. For force feedback, the host controller solves the joint torque relations based on the kinematics of the masterarm and the satellite controller controls each joint's torque. This distributed controller greatly reduces calculation and control loads and makes the electrical wiring very simple. In order to measure finger movement of the operator, the master hand is developed using infrared light. A similar slave satellite controller for this master hand is interfaced to the host controller via a SPI network. For teleoperation of the KIST humanoid robot which has two arms and two hands, the masterarm controller with one host controller and 16 satellite controllers is implemented. Force feedback with less than 2 msec update rate is achieved. This masterarm/hand is also integrated with a graphic model instead of using the slave robot. The experimental results show the performance of the distributed controller.

Force guided robot control is a control scheme based on the interpretation of measured force acting on the robot end effector. A functional map relating the correction of motion to force measurements is generated based on the geometry of the workpiece and its kinematic behavior in interacting with the environment. In the traditional force guided control schemes, the contact force measured by a force sensor is directly fed back to a feedback controller to generate a motion correction signal. In the paper, instead of simply measuring contact forces, we take positive actions by giving perturbation to the end effector and observing the reaction forces to the perturbation in order to obtain much richer and more reliable information. By the correlation between the input perturbation and the resultant reaction forces, we can determine the gradient of the force profile and guide the part correctly. By applying a type of direct adaptive control, the contact force is maintained at the lowest level. This algorithm is applied to a pipe insertion task, in which the insertion force is minimized during the insertion. Based on the process model and stability analysis using the Popov stability criterion, conditions for stable, successful insertion despite nonlinearities and uncertainties in the environment are obtained. The theoretical results are verified using the experimental data. To generate high frequency perturbation, a vibratory end effector using piezoelectric actuators is designed and built. Through both simulations and experiments, the feasibility and usefulness of these methods are demonstrated.

A lot of researches have been done in the teleoperation field. In order to control a robot from remote site, simple position teaching devices were developed followed by the advanced force feedback devices. Most of the force feedback master devices are as big as the slave robot and equipped with heavy actuators. A new teleoperation device is presented in this paper. A simple exo-skeleton type master arm is designed based on the kinematic analysis of the human arm. For force reflection, pneumatic actuators are used. This device is integrated with KIST Humanoid robot as well as graphic simulator, so that the master arm can be used for teaching the robot or as the virtual reality device integrated with the graphic simulator. A force reflecting master hand was also developed with similar design to the master arm. These force reflecting master arm and hand are very light and compact so that a human can easily wear while he feels the same force as the robot does while giving position command to the slave robot.

To increase the accuracy and the reliability of teleoperation, close interaction between human and the slave robot is required. Korea Institute of Science and Technology (KIST) is developing a Humanoid robot, which has two 9-DOF arms, two 4-DOF hands, a 3-DOF waist and a 2-DOF neck. An automatic path planner which takes advantage of redundancy is also developed, but it requires huge computational power and time. For tasks which are very complex, or interacting with environments, robot path can be generated very easily and accurately by human teaching, hence the automatic path planner is not required. Using sensors, human motion is analyzed and converted to robot joint angles. For tasks interacting with unknown environment, teleoperation with force feedback is more useful. A simple platform-type master arm is designed based on the kinematic analysis of human. For force reflection, the pneumatic actuators are used. This device is integrated with KIST Humanoid robot as well as graphic simulator, so that the master arm can be used for teaching the robot or as the virtual reality device integrated with the graphic simulator. Force reflecting master hand is also developed. These force reflecting master arm and hand are very light and compact so that a human can wear easily while he feels the same force as the robot does while giving position command to the slave robot. In addition to master arm/hand, visual information is feedback to human through HMD for higher interaction. Therefore, a new way of integration of human and the robot is developed.

A lot of research has been done in robot path generation, including automatic path planning and teaching/playback. Teleoperation is commonly used to control a robot from a remote site in an unknown environment. Korea Institute of Science and Technology (KIST) is developing a Humanoid robot, which has two arms (9 degrees of freedom for each), two hands (4 d.o.f. each), a waist (3 d.o.f.) and a neck (2 d.o.f.) An automatic path planner which takes advantage of redundancy is also developed, however, it requires huge computational power and time. For some tasks, which are very complex, or interacting with environments, the robot path can be generated very easily and accurately by human teaching, hence the automatic path planner is not required. Using sensors, human motion is analyzed and converted to robot joint angles. For tasks interacting with an unknown environment, teleoperation with force feedback is more useful. A simple exoskeleton-type master arm is designed based on kinematic analysis. For force reflection, pneumatic actuators are used. This device is integrated with KIST Humanoid robots. A force reflecting master hand is also developed. In addition to the master arm/hand, visual information is fed back to the human through HMD for higher interaction, and voice command interpreter/voice synthesis is integrated. Therefore a new method of human/robot integration is developed.

An approach for overcoming the difficulties in force-guided assembly for real manufacturing applications is presented. The perturbation/correlation based control for a robot system is generalized, and then applied to the pipe insertion task which is difficult using conventional force guided control. Based on the analytic model of this controller, a Popov stability criterion is applied to cope with the unknown nonlinearity in the environment. Several critical assumptions are verified based on the experimental data, and the guide line to select parameters involved in this method is presented. Through an experiment, the feasibility and usefulness of this approach are demonstrated. A vibratory end-effector using a piezoelectric actuator was designed and implemented.

In the traditional force-guided control schemes, the contact force measured by a force sensor is directly fed back to a feedback controller to generate a motion correction signal. The issue central to force guided robot control is how to obtain reliable, consistent and copious force signals and extract useful information in order to successfully guide the robot while keeping the contact force at a desired level. In this paper, instead of simply measuring contact forces, we take positive actions by giving perturbation to the end effector and observing the reaction forces to the perturbation in order to obtain much richer and more reliable information. By taking the correlation between the input perturbation and the resultant reaction forces, we can determine the gradient of the force profile and guide the part correctly. This algorithm is applied to a pipe insertion task, and connector assembly task. Based on the process model and stability analysis using the Popov stability criterion, conditions for stable, successful insertion despite nonlinearities and uncertainties in the environment are obtained. The theoretical results are verified using the experimental data.

This paper presents a motion planner for robots in a game situation, in which the agents involved have conflicting goals. More specifically, a human-computer boxing game system is developed in a graphical environment. One boxer is controlled by a human through a motion tracking system, while the other is controlled by a motion planning algorithm. The planner has a defensive and an offensive strategies. The defensive strategy tries to move one of the forearms between the opponent's fist and its body, while the offensive strategy locates an exposed part of the opponent's body and throws a punch at it. To achieve an interactive operation, the motion planner considers only kinematic and geometric constraints during the search. This project is also aimed at integrating a planner with all actual robot, and investigating associated technological bottlenecks.

A new approach to force guided assembly is developed for the assembly of unsurfaced parts having sharp burrs and irregular surfaces. Due to friction at burrs and irregular surfaces, force signals are very noisy and erratic, preventing reliable sensing and monitoring of the assembly process. In this paper, instead of simply measuring contact forces, we take positive actions by actively shaking the end effector and observing the reaction forces to the perturbation in order to obtain rich, reliable information. By taking the correlation between the input perturbation and the resultant reaction forces, we can determine the direction of the part surface and guide the hand-held part correctly despite burrs and poor surface finish. The principle of active force sensing using a correlation technique is described. An algorithm for guiding an assembly part by using the correlation information is developed based on the theory of direct adaptive control. The method is then applied to a practical task: the assembly of sheet metal parts with sharp edges and burrs. Experiments demonstrate the effectiveness and feasibility of the active force sensing method.>