Mapless Exploration Navigation Based On Reinforcement Learning

With the development of robotics,mobile robots are required to automatically explore and complete mission in unknown areas.Compared with traditional methods,a new generation of deep learning algorithm based on deep learning algorithm can complete exploration navigation tasks without relying on priori map information even in complex and unknown environments.Based on deep reinforcement learning,I studied the path planning of mobile robots and explore navigation technology.And I mainly studied the depth Q learning algorithm based on value function,the actor-Critic-based A3 C algorithm and the internal drive technology which can encourage robots to actively explore.The main contents of the paper are as follows:First,the algorithm model of time difference is analyzed.The model is used as the sampling model.The deep Q learning algorithm is studied and implemented.Aiming to solve the problem of the feature of the map environment may be more complicated,I use the experience playback technology and fixed network model technology to enhance the network which improve the network convergence ability.The effectiveness of the algorithm is verified by experiments.Second,after clarifying that the exploring map is very complicated,it is difficult to train deep Q model and the convergence is slow.The Actor-Critic algorithm which combines deep Q learning and policy-based reinforcement learning algorithm is studied which is named AC.Select the A3 C model from AC model A2 C model and A3 C model,which is explored by multiple computer threads and summarizes the results.The algorithm combines the advantages of deep Q learning and policy-based algorithms in single-step training.At the same time,the policy probability can also be output.In the end,the convergence and stability of training are far superior to the depth Q learning algorithm.Finally,in order to enhance the exploration ability of mobile robots,research and join the internal driving mechanism that imitates the human curiosity,and encourage the mobile robots to explore more in the initial stage instead of being limited to the developed parts.The argument proves that in the same kind of algorithm,adding the internal drive to the reward of the mobile robot can make the mobile robot realize the map full coverage faster.