Path Planning of Mobile Robots using Deep Reinforcement Learning

Computer and Knowledge Engineering

مقالات آماده انتشار، پذیرفته شده، انتشار آنلاین از تاریخ 11 آذر 1399  

نوع مقاله: Machine learning-Sadoghi

شناسه دیجیتال (DOI): 10.22067/cke.2020.39287

نویسندگان

Nematollah Ab azar ؛ Aref Shahmansoorian؛ Mohsen Davoudi

Imam Khomeini International University

چکیده

This article approaches learning the path planning in mobile robots using reinforcement learning methods including Q-learning, SARSA,  and multi-step-ahead prediction  combined with Deep Learning method. The motivation of the authors is to formulate and solve the optimization problem of learning path planning considering avoiding obstacles and path optimality. In this regard, a reward function was constructed based on the weighted features of the environment constraints and path optimality criteria. The related algorithms have been simulated and the obtained data have been analyzed in a comparative framework. The results illustrate that using deep reinforcement learning including Deep Q-learning, Deep SARSA and Deep  in mobile robot path planning leads to finding optimal path 5-10 % faster and better than pure reinforcement learning methods.