Learning structured reactive navigation plans from executing MDP navigation policies

Abstract

Autonomous robots, such as robot office couriers, need navigation routines that support flexible task execution and effective action planning. This paper describes \xfl, a system that learns structured symbolic navigation plans. Given a navigation task, \xfl\ learns to structure continuous navigation behavior and represents the learned structure as compact and transparent plans. The structured plans are obtained by starting with monolithical default plans that are optimized for average performance and adding subplans to improve the navigation performance for the given task. Compactness is achieved by incorporating only subplans that achieve significant performance gains. The resulting plans support action planning and opportunistic task execution. \xfl\ is implemented and extensively evaluated on an autonomous mobile robot.