Autonomous motion planning using a predictive temporal method .

摘要

The introduction of moving obstacles into a robot's environment presents added complexity to the motion planning task. This dissertation examines the need for and development of a representation which incorporates the dynamic nature of the environment and presents a novel motion planning method which utilizes this representation to facilitate the generation of optimal trajectories among moving obstacles, termed the predictive temporal motion planning (PTMP) method. This new method provides an advanced approach to the problem of generating solution trajectories in dynamic environments by elegantly connecting the tasks of obstacle detection and prediction, environment mapping and motion planning.The dynamic environmental representation takes the form of a typical grid which is extended into the time dimension by adding temporal layers to the grid structure. The layers of this temporal grid represent distinct time-steps into the future. These time-steps are determined by considering how the motion planning algorithm calculates its discrete control commands. Obstacle motion prediction is incorporated into the temporal grid by estimating future positions of moving obstacles and displaying these estimates in the layer of the temporal grid associated with the prediction times.The new motion planning method then can use this predictive temporal grid to investigate potential control input sequences to generate an optimal trajectory to achieve its goal. As the algorithm evaluates potential control commands at various time-steps in the future, it does so by exploring the various temporal layers of the new grid structure corresponding to these distinct control times. By considering the estimated future motions of any obstacles, the motion planning algorithm can more intelligently calculate its control sequences to avoid these objects in an efficient manner.The research presented covers the theory of this new method and a specific implementation on an unmanned ground vehicle platform at the University of Florida's Center for Intelligent Machines and Robotics (CIMAR). Results from simulation have shown that the PTMP method is viable and advantageous for the motion planning required by robotic systems in dynamic environments. (Full text of this dissertation may be available via the University of Florida Libraries web site. Please check http://www.uflib.ufl.edu/etd.html)