Robot motion planning integrating planning strategies and learning methods

Abstract

Robot motidn planning in a dynamic ch, ttered workspace requires the capability of dealing with obstacles and deadlock situations. The paper analyzes situations where the robot is considered with its shape and size and it ca,, only perceive the space through its local sensors. The robot explores the space using a pla, aer ba~ed on an artificial potential field and incrememally learns a fast way to escape or preve,~t deadlock situations using a combiuation of sensor percep-tions, field information and planner experience. The knowledge acquired is a high-hvel network useful for avoiding deadlock areas consisting of lo-cal minimum nodes, backtracking nodes a,Ld sub-goal nodes.