Research On Path Planning For Mobile Robot Based On State Prediction Reinforcement Learning

With the development of robot technology, it has begun to be applied to unknown environment now. Compared with the research on the path planning in static known environment, the unknown environment and dynamic changes bring new challenges to the path planning of environment exploration for mobile robot. Ineluctably, mobile robot will meet various obstacle because of unknown environment when exploring. Therefore, the mobile robot which can obstacle avoidance and vivid programming in unknown environment has important practical significance. In this paper, we use reinforcement learning to study the path planning for mobile robot both in static and dynamic environment.Firstly, this dissertation sums up the research on path planning for mobile robot exploration based on reinforcement learning.Then reviews the research and development about path planning of mobile robot environment exploration.The background and main contents of this dissertation are described briefly.Secondly, this part introduces the relevant knowledge, present condition and existent problem which mobile robot environment exploration in detailed, include established of grid maps, the cost of reaching a target point and its utility, the distribution of targets for multiple robots etc.. Then expatiation the method of path planning, sense system and the conflict resolution of multi-mobile robot.The third part introduces the concept, principle, method, algorithm and the research of present condition about reinforcement learning in detailed.Then, aiming at the static environment exploration of single robot, the key to described the path planning strategy based on Q reinforcement learning by dividing the state and act space, structuring of reinforcement function etc..The fourth part uses the reinforcement learning and the thought of prediction at single robot path planning in dynamic environment,that in order to solve the problem of obstacle avoidance. Because of the each previous step decision to the success or failure, our approach lead a technique of Eligibility Trace and use the improvement of Q reinforcement learning algorithm to carry out the control.The fifth part of this dissertation learns from mankind who make sure through the prediction, and combines the state prediction with the reinforcement learning uses for multi-mobile robot system on path planning.The approach is more reasonable than the method which only use reinforcement learning algorithm, and that carry out the reasonable collide avoids between the robot, in order to lower a space size and raise calculation speed.Finally, conclusions are given with recommendation for future work.