China University of Mining and Technology Xuzhou, 221116, China

---

Abstract—Along with the development of robot technology, a robot not only need to complete a specific task, but aslo need to do path planning in the process of performing the task. So, path planning is widly studied. This paper introduce a method of robot path planning based on reinforcement learning,which aimed at Markovian decision process. In this paper, we introduce the basic concept, principle and the method of reinforcement learning and some other algorithms.Then,we do research from single robot’s path planning in the static invironment based on Qlearning, and describe the application of this algorithm on the path planning by setting off state space and action space reasonablly and designing reinforcement function.By edditting Matlab program,we do some simulation experiments,which incarnate the algorithm visually and get the optimal path.

Index Terms—reinforcement learning, markov decision process, Q-learning

[PDF]

Cite: Qian Zhang, Ming Li, Xuesong Wang, and Yong Zhang, "Reinforcement Learning in Robot Path Optimization," Journal of Software vol. 7, no. 3, pp. 657-662, 2012.