Efficient Symbolic Task Planning for Multiple Mobile Robots
University of Texas at Austin Austin United States
Pagination or Media Count:
Symbolic task planning enables a robot to make high-level decisions toward a complex goal by computing a sequence of actions with minimum expected costs. This thesis builds on a single-robot planning framework, and aims to address two issues 1 lack of performance in-formation in the selection of planners across different formalisms, and2 time complexity of optimal planning for multiple mobile robots. In this thesis we first investigate the performance of the state-of-the-art solvers of Planning Domain Definition Language PDDL and Answer Set Programming ASP in robot navigation problems. We then aim to reduce overall costs where multiple mobile robots may block in narrow corridors or collaborate to open doors. It is challenging to model such interactions due to uncertain delays of navigation actions in populated areas. This paper addresses this challenge with an algorithm which calculates the conditional distribution of plan costs for each robot, given planned actions of other robots. We then propose an iterative conditional planning algorithm to efficiently approximate optimal plans of the system. Experiments in simulation and a demonstration on real robots show that the algorithm has a significant advantage over baselines in which robots plan individually, or plan together without a model for navigation action durations.