Least mean p-power extreme learning machine for obstacle avoidance of a mobile robot

Abstract

This paper proposes an obstacle avoidance method for navigation of a mobile robot in uncertain environments based on a novel neural learning algorithm, namely least mean p-norm extreme learning machine (LMP-ELM) and Q-learning. The proposed obstacle avoidance method comprises of two behavior modules, Viz., an avoidance behavior and goal-seeking behavior. At the learning phase, the two modules are independently designed using the proposed LMP-ELM and Q-Learning. And then they are combined to navigate the mobile to the goal position without colliding with obstacles based on a switching function at the running phase. The LMP-ELM is used to realize the state-action mapping of the Q-learning. In the novel LMP-ELM, the computationally simple extreme learning machine architecture is maintained but a novel error criterion, namely the least mean p-power (LMP) error criterion provides a mechanism to update the output weights sequentially. The LMP error criterion aims to minimize the mean p-power of the error that is the generalization of the mean square error criterion used in the ELM. The effectiveness of the proposed method is verified by a series of simulations.