Fault-Tolernat Vision-Guided Robotic Systems for Aerospace Applications

点击翻译按钮获取中文摘要

Space robotics manifested by the iconic Canadarm technology is important to Canadian national interest for scientific, commercial, and strategic reasons. In particular, reliable vision-guided systems are critical for autonomous operation in many current and near future robotics space missions to support rendezvous, proximity robotic operations, and automated planetary landing. Adoption of aerial robotic vehicles equipped with vision-guided system for a variety of civilian uses is also important for Canada due to the vastness of its domestic airspace. Despite of significant progress made in the past two decades, vision guided robotic systems still face many challenging problems mainly due to undependability of vision systems, environmental uncertainties, and practical GN&C given multiple physical and operational constraints. Testing vision-guided robotics systems in aerospace applications is particularly challenging because the tests often should perform in a laboratory environment whereas the eventual vision-guided robotic systems operate in an aerospace environment.This research program is aimed at enhancing the robustness of vision-guided robotic system through a series of advancements in fault detection and recovery, adaptive supervisory control, as well as scalable hardware-in-the-loop simulation test methods. The methodology is general enough for applying to space and aerial robotic systems alike. The ultimate goal is to develop an aerospace robotic system cable of adaptively tuning itself against not only inaccurate and potentially erroneous visual information but against the dynamics and calibration uncertainties which affect the flight dynamics and system performance. The adaptive supervisory control chooses the most appropriate control action if partial or complete failure of the vision system happen. Development of a scalable hardware-in-the-loop simulation for realistic testing of vision-guided robotic systems is also aimed at in this research. Real-time 3D vision data to feed the vision algorithm and GN&C are generated by an actual laser range finder or a stereo camera, while the relative motion in a proximity operation or planetary landing is generated by a simulating robot according to flight dynamics. A dimensionless mathematical technique, analogous to fluid mechanics/dimensional analysis, will be adopted for proper scaling of the mockup geometry and the simulated states such as range and velocities in order to achieve dynamic similarities in the face of physical limitations of the laboratory testing.The graduate students involved in this proposed research will have a unique opportunity to acquire practical skills in aerospace engineering, while working on problems at the cutting edge of the theoretical development of advanced GN&C, adaptive estimation, 3D vision systems, fault recovery strategies, and HIL simulation technology.