Model Predictive Control Strategies for Mixed Telerobotic/Autonomous Robotic Systems

点击翻译按钮获取中文摘要

In recent years, technological and scientific advances in a number of disciplines such as mechatronics, manufacturing, information and signal processing, controls, and artificial intelligence have fuelled a rapid growth in traditional and new applications of robotics. Robotic systems are increasingly used in industrial automation, disaster recovery, search and rescue, hazardous material and waste handling, mining, space and underwater operations, and medicine. Notwithstanding all these advances, robots may not still be able to operate fully autonomously in many complex uncertain task environments, and do often require some form of human supervision/control. This research is concerned with human-in-the-loop robotic systems. These are systems in which human(s) and robot(s) work cooperatively to accomplish a task. The overall goal of the research is to bridge an existing gap between the two fields of autonomous robotics and telerobotics. Traditionally, research in these areas has followed two mostly separate paths. Autonomous robotics systems have mostly comprised of fully autonomous robots. Telerobotics systems, on other hand, have mainly involved system configurations in which one operator fully controls one robotic manipulator. This research pursues a new paradigm in system design and control in which elements of telerobotics and autonomous robotics are combined. In this new paradigm, the operator(s) control aspects of the task that would benefit from human’s unique cognition and decision-making capabilities. Meanwhile, robot(s) assist the operator(s) by autonomously controlling more structured aspects of the task to improve precision and reduce cognitive load. The proposed research will seek a general framework for control and coordination in mixed autonomous robotics/teleorobotics, applicable to a broad class of system configurations. The control strategies will be based on a novel multiple-level control architecture. At one level, optimization-based model-predictive controllers that deal with autonomous aspects of the task will produce commends, which will then be passed to another level of control. This next level controller will combine autonomous and conventional teleoperation control and will resolve any potential conflicts between the two based on user-defined task priorities. Uncertainty in operator(s) future actions and other elements of the task environment will be considered in making optimal control decisions. This will be achieved using particle-based estimation and optimization techniques. It is expected that the real-time computational requirements of the resulting estimation and control algorithms will exceed capabilities of state-of-art desktop computers. Parallel implementation of the algorithms on Graphic Processor Units will be pursued to help achieve the timing requirements using inexpensive, off-the-shelf, graphics cards. The outcomes of this research will help robotics engineers and scientists to design systems that combine benefits of teleoperation and autonomous control. Such systems will be able to operate more effectively in complex unstructured environments, which are common in many robotics applications. The new knowledge arising from this research will find its way into new technologies and products in areas of great importance to Canada such as healthcare, manufacturing and auto industry, mining, and space. Canada is a major player in the field of robotics and Canadian companies in this area will benefit from the new knowledge and also the highly qualified personnel that will participate in the research.