Autonomous exploration for mobile robot using Q-learning

Abstract

This paper proposes a reinforcement Learning network based on an OS-ELM (online sequential extreme learning machine) to realize the path planning function of mobile robot in the indoor environment. Through Q-learning, a model-free reinforcement learning technique, the mobile robot can autonomously search the target to traverse the room and learn to train obstacle avoidance function in an absolutely strange environment. The system is based on the Q-OS-ELM network. Instead of using traditional back propagation neural network, ELM could remarkably improve the efficiency and speed of Q function approximation. Compared with the traditional ELM (extreme learning machine) network, OS-ELM can update the output weight in real time, which greatly reduces the stacking and redundancy of the samples and improves the efficiency of Q-network.