Real-time dynamic trajectory optimization with application to free-flying space robots.

摘要

The capability of robots to complete tasks or entire missions autonomously relies heavily on their ability to plan. Good planners must not only be able to produce efficient plans but must also be able to modify those plans quickly in response to unpredicted events. Unfortunately these two goals are often conflicting ones, with only slow, complex planners able to produce efficient plans, and only quick, simple planners able to react to unpredicted events.;The research presented in this dissertation focuses on the development of a real-time dynamic trajectory optimization system that provides both highly efficient motion planning capabilities and the ability to react to uncertainty in the environment. This system achieves these capabilities by utilizing simultaneous planning and execution to improve the robot's trajectory while the robot is in motion along the trajectory. Although other robotic systems have used simultaneous planning and execution, this dissertation applies the concept to dynamic trajectory optimization, a sophisticated technique for computing highly efficient trajectories that has previously been used only in off-line planning applications and in simulation.;The resulting system uses a non-linear optimization algorithm to improve an initial trajectory, subject to the dynamics of the system and constraints on the robot's motion, in order to minimize a weighted sum of the fuel and time required to complete the trajectory. Using this system, several motion planning tasks are demonstrated experimentally on a thruster propelled free-flying robot. The most complex of these tasks requires the robot to travel around a pair of stationary obstacles and intercept a moving, maneuvering target vehicle in a highly efficient manner. The experimental results show that for the sample moving-target-intercept task, the real-time planner provides 2.42 times better performance than a reactive planner and an on-line planner is unable to complete the task at all. This experimental demonstration highlights the advantages of real-time dynamic trajectory optimization in providing a high performance motion planning capability, even when operating in a dynamic, uncertain environment.