Improving Prehensile Mobile Manipulation Performance through Experience Reuse

Abstract

During pick and place tasks, a mobile manipulator performs recurring relative moves within the close proximities of the object of interest and the destination independent of their global poses. These moves are usually critical to the success of the manipulation attempt and hence need to be executed delicately. Considering the critical yet recurring nature of these moves, we let the robot memorize them as state-action sequences and reuse them whenever possible to guide manipulation planning and execution. When combined with a sampling-based generative planner, this guidance helps reduce planning time by deliberately biasing the planning process towards the feasible sequences. Additionally, monitoring the execution while reiterating the reached sequences improves the task success rate. Our experiments show that this complementary combination of the already available partial plans and executions with those generated from scratch yields fast, reliable, and repeatable solutions.