Feature Space Decomposition for Autonomous Robot Adaptation in Programming by Demonstration

Adaptation is an essential capability for intelligent robots to work in human environments. We propose a novel Feature Space Decomposition (FSD) approach to effectively address the robot adaptation problem, which is directly applicable to the learning framework based on Programming by Demonstration (PbD) and Reinforcement Learning (RL). Our FSD method decomposes the high-dimensional original features into principal and non-principal feature spaces, which are respectively used to learn a generalized trajectory from demonstrations, and to form a new low-dimensional search space on RL for autonomous robot adaptation beyond demonstrations. Experimental results on real robots validate that our FSD approach enables the robots to effectively adapt to new environments, and is usually able to find optimal solutions more quickly than traditional approaches when significant environment changes occur.