A Reinforcement Learning with Adaptive State Space Construction for Mobile Robot Navigation

摘要

Reinforcement learning provides a framework for the navigation problem of mobile robot in unknown environment. However, the generalization ability and learning efficiency of RL-based navigation systems have to be improved to satisfy the requirements of the continuous sensory information of mobile robots. In this paper, we propose an adaptive state space construction strategy for reinforcement learning based on competitive neural network. The method can adjust the size of the state space appropriately according to the task complexity and progress of learning and overcome the difficulty of dimensionality cruse. The simulation results are provided to demonstrate the validity of the proposed method in solving mobile robot navigation..