Research On Path Planning Algorithm Of Mobile Robot Based On OpenAI Gym And DRL

Mobile robot path planning technology,that is,to find an effective path that meets the constraints between a given starting point and a target point(constraints can be no collision,shortest path,low energy consumption,etc.),as an important technical support for robotics,has received more and more attention and research.Reinforcement learning,as a machine algorithm that searches for the optimal strategy through continuous "trial and error" interaction with the environment,has excellent versatility and effectiveness.This article combines its excellent characteristics to propose an efficient path planning algorithm for mobile robots.Combining the actual needs of mobile robots,the reinforcement learning algorithm is improved.Finally,a simulation experiment is carried out in a simulation environment to verify the superiority of the improved algorithm.The specific research content is as follows:1.Summarize the current research status of the path planning problem of mobile robots,analyze the advantages and disadvantages of using traditional algorithms,intelligent algorithms,graph theory-based methods and machine learning algorithms in solving path planning problems,and comprehensively consider the working environment of mobile robots.Demand,choose the more popular reinforcement learning algorithm in the field of machine learning for discussion and research.2.The more classic Q-learning algorithm in the field of reinforcement learning is selected for research.When the Q-learning algorithm is used to solve the single robot path planning problem actually,it is found that there are three problems in Q-learning.In response to these three problems,the author made corresponding improvements to the Q-learning algorithm,which effectively improved the work efficiency and performance of the Q-learning algorithm.3.In order to apply the single-robot Q-learning path planning algorithm to the multi-robot field,combined with deep learning ideas,the DQN version of multi-robot coordinated obstacle avoidance is realized;and the method of introducing prior knowledge and prior rules is adopted to improve Algorithm performance and reduce training overhead.Finally,in the simulation experiment,the path planning of multi-robot coordinated obstacle avoidance was realized.4.In order to further improve the efficiency of multi-robot path planning,this paper combines the attention mechanism of the deep learning field to give similar visual attention to the strategy agent,and improves the experience playback mechanism,and proposes a time-based priority experience playback method,Enabling the agent to pay more attention to the more important parts,accelerating and optimizing the model training process.5.A simulated labyrinth training environment is constructed based on the openAI gym tool.Experiments have proved that the improved algorithm proposed in this paper performs well in dealing with the path planning problem of mobile robots.