Robot Planning in Uncertain, Dynamic Environments

Abstract

Many real-world applications require robots to operate in dynamic environments characterized by moving objects or agents whose trajectories are unpredictable. This thesis addresses the challenges posed by such environments by introducing Relative Temporal Probabilistic Roadmaps (Rel-T-PRM), a novel motion planning algorithm that builds upon the Temporal Probabilistic Roadmap (T-PRM) algorithm. The Rel-T-PRM allows for variable dynamic obstacle size, enables robustness with respect to minor changes in time and position and and introduces the concept of waiting until obstacles clear. Furthermore, we leverage Rel-T-PRM’s strengths to propose two replanning strategies. The first attempts to rapidly replan on-the-fly by using waiting to modify the trajectory without needing to modify the path. The second proposed replanning strategy identifies and plans to safe locations, where the robot can safely replan under a longer time horizon. We demonstrate Rel-T-PRM through a variety of simulation experiments on a fixed-base robotic manipulator.