Internet-based teleoperation for robot navigation

Abstract

Internet-based robot teleoperation obviates the need for dedicated networks and devices, reduces costs, extends operating distances, and allows precious resources sharing for public education or academic research. Except for operating in hazardous environments, Internet telerobotics has opened up a new range of real-world applications, involving tele-manufacturing, tele-training, tele-surgery, museum guide, space exploration, disaster rescue, and health care. There are many problems on Internet-based teleoperation that need to be addressed, such as data transmission over uncertain time-delay and unreliable Internet, teleoperation by inexperienced users, short of interactivity, and so on. Moreover, Internet robots require a much higher degree of autonomy than traditional teleoperation so that the robots are able to ensure safe operations and perform some tasks autonomously. In this thesis, we aim at developing a practical robotic system for the target application: the inexperienced Internet users can remotely control a wheeled robot which is able to perform some complex tasks autonomously (e.g. active map learning, goal-oriented navigation) or to interact with human operator in order to explore unknown and dynamic environments. The experiments are based on a Pioneer robot that is equipped with an onboard camera and eight forward ultrasonic sensors. The control commands transfer through radio Ethernet devices. To help realize such robotic system, the research is conducted on the following aspects: 1) The video transmission via the low-bandwidth Internet is investigated and implemented so that the robot's surroundings can be seen by any remote operators through the images captured from an onboard camera. It is a prerequisite to develop a practical teleoperation system. Traditional approach is via the picture transmission (e.g. JPEG or GIF), which leads to a very poor quality of service (QoS) because of the high latency of the Internet, such as long time delay, data error or restricted bandwidth. The thesis investigates and develops a streaming technology based approach for streaming video transmission. Two video compression algorithms (WMV9 and MPEG4) under different bandwidth, two video encoding methods (CBR and Quality-based VBR) as well as the transmission stability and time delay have been investigated. 2) A framework for autonomous navigation using fuzzy logic is proposed. This work is a base for the subsequent designs of intelligent control programs so that the mobile robot is able to autonomously perform some complex tasks amid various degrees of uncertainties. The proposed framework involves goal determination, preprocessing, behavior design, behavior arbitration, and command fusion. Traditional framework for autonomous navigation is SMPA (Sense-Model-Plan-Act) approach, which is inadequate for dealing with unknown and dynamic real world. The behavior-based approach can act in real-time and has good robustness in such environments. The preprocessing module is used to reduce the complexity of input space by introducing a limited number of intermediate variables. The elementary behavior can be designed using fuzzy logic controller or an analytic algorithm. A behavior arbitration module is used to calculate the crisp weighting factors of each elementary behavior. The final robot motion output is obtained by the command fusion for a weighting combination of all elementary behaviors. A goal-oriented navigation task, combined with obstacle-avoidance (OA) and goal-seeking (GS) behaviors, is implemented as an example of the proposed framework. 3) A new teleoperation approach so called telecommanding is proposed to provide an interactive control interface and a complete framework for control management and command processing. The traditional direct control reduces the stability of control loop because the controlled robot has no local intelligence and it needs to maintain continuous connection. The existing supervisory control methods are inadequate mainly in that they fail to provide human-robot interactivity. The proposed approach involves two different but complementary commands: joystick command (e.g. LEFT, RIGHT, UP, and DOWN) and linguistic command (e.g. MOVE, TURN, GOTOEND, WANDER, COORDINATE, and MAPPING). Each command is designed to perform independent task, which is defined with multiple events (non-time action references) and corresponding response functions. Simulated and real world experiments have been conducted to test the use of both joystick commands and linguistic commands for Internet-based robot teleoperation. The advantages and disadvantages as well as stability of telecommanding are analyzed. 4) To model a priori unknown environment (i.e. a MAPPING linguistic command), a new map learning approach called memory grid mapping is proposed. The robot builds a map based on robot's sensory information and actively explores the unknown environment. The approach includes a map model, a map update method, an exploration method, and a map postprocessing method. The map adopts a grid-based representation. A so-called obstacle memory dot (OMD) matrix is designed to save the frequency values which measure the confidence that a cell is occupied by an obstacle. A so-called trajectory memory dot (TMD) matrix is designed to save the trajectory traversed by the robot in order to facilitate the online path planning. Two behaviors, path-exploring behavior and environment-detecting behavior, are coordinated to make the robot exploring a least known environment. The map postprocessing method includes a threshold operation, a template operation, and an insert operation. The efficiency of map learning is investigated. The map accuracy under different cell sizes and different map postprocessing is investigated as well. Experiments are done for the map learning in different simulation environments. 5) For a teleoperated mobile robot that is exploring unknown indoor environments, it is desired that the robot is able to autonomously arrive at a given goal location (i.e. an enhanced COORDINATE linguistic command), even though the environments involve all kinds of complex situations with local minima. The thesis proposes a new navigation method, namely minimum risk method, to realize such function. The method makes use of the proposed memory grid map. When a mobile robot is performing the goal-oriented navigation, it updates a memory grid map in real-time. A novel path-searching (PS) behavior is developed to use the map information and to recommend a safest regional direction that can enable the robot to detect potential local minima and escape from them. The final command outputs are obtained by coordinating the behaviors: PS, OA, and GS. Fuzzy logic controllers are used to implement behavior design and coordination. The method is experimentally demonstrated to give global convergence to a given goal location, even though it is used in the long-wall, large concave, recursive U-shape, unstructured, cluttered, maze-like, or dynamic (i.e. with moving human) environments. The developed telerobotic system has been demonstrated to be feasible to provide the service of Internet-based teleoperation in university campus and exhibition center. The tests have been performed successfully through the Internet remotely from overseas places (e.g. Canada, Singapore, Chinese Beijing, Shanghai, Xiamen) to Hong Kong.