| As a branch of intelligent robotics,mobile robots are increasingly being applied to people’s production and daily life.Path planning,as one of the key technologies of autonomous navigation of mobile robots,has always been a research hotspot for scholars at home and abroad.With the rapid development of artificial intelligence technology and the continuous improvement of computing power,path planning is combined with more and more intelligent algorithms and applied to various fields.Traditional mobile robot path planning algorithms require prior knowledge of the environment to plan a global path on a known map,lacking the ability to explore paths autonomously and find a collision-free path from the starting point to the destination in an unknown environment.Therefore,this paper combines deep reinforcement learning methods with path planning,explores and learns through the interaction between the robot and the environment,continuously optimizes the robot’s decision-making ability,and ultimately achieves path planning and obstacle avoidance in complex environments.The main research content of this paper is as follows:First,the principle of deep reinforcement learning algorithm is explained,and the deep Qnetwork(DQN)algorithm used in this paper is introduced in detail,and improvements are made to the shortcomings of the DQN algorithm.By replacing the exploration strategy of the traditional algorithm with the adaptive exploration strategy,the problem of low exploration efficiency of the agent in the initial training is solved.After that,the multi-step guide method and preferential experience replay mechanism are introduced to improve the algorithm performance in order to solve the problems of slow convergence speed and low sample utilization rate in the DQN algorithm.Second,in order to better solve the path planning problem of mobile robots,the kinematic model of mobile robots is established,and a path planning algorithm framework based on improved DQN is constructed according to the principle of reinforcement learning.The state space,action space,and reward function in the reinforcement learning method model are designed,and the problem of sparse rewards in the traditional reward function is improved.Finally,the path planning algorithm proposed in this paper is experimentally tested to verify its effectiveness and feasibility.Three different grid maps were built using the Tkinter library in Python,and comparative experiments were conducted on the traditional DQN algorithm and the improved DQN algorithm.The experimental results show that the improved DQN algorithm improves learning efficiency,speeds up convergence speed,and produces better path planning results.To make the algorithm more realistic,a simulation environment was built on the ROS Gazebo platform under the Linux system to verify the improved DQN algorithm.It was found that the improved DQN algorithm can perform well in path planning tasks in unknown environments. |
