Intelligent Car Path Planning Based On Reinforcement Learning

With the development of intelligent robots research,intelligent cars,as an important branch of intelligent robots,play an increasingly important role in all aspects of human life.For intelligent cars,the ability to plan a path and avoid obstacles in the environment is one of the most basic tasks.At present,intelligent cars are widely used in unknown environments.Compared with the known environmental conditions,this greatly increases the difficulty of path planning and brings great challenges to the work of researchers.Therefore,it is very important to make the intelligent cars self-adaptive path planning and obstacle avoidance in an unknown environment.Because reinforcement learning is a machine learning algorithm that enables an agent to continuously interact with the environment without prior knowledge,thereby adjusting the action strategy and finally learning the optimal action strategy,it is often used to solve the path planning problem of the intelligent car in the unknown environment.However,there are several shortcomings in reinforcement learning that cannot be ignored.When the environment becomes complicated,it will produce a ” curse of dimensionality ”.At this time,the convergence of the algorithm is worse and it takes a long time to learn;and the model obtained by reinforcement learning has poor generalization,it cannot adapt well to other unknown environments.Therefore,the research content of this paper is to optimize reinforcement learning,speed up its convergence,and improve the adaptability and generalization of the model.In this paper,we develop deep Q-learning based on heuristic knowledge and experience replay.And we have simulated and compared our method with other methods.The simulation results show that in comparison with other methods,our method can converge to an optimal action strategy with less time and can explore a path in an unknown environment with fewer steps and larger average reward.The main research work in this paper is as follows:(1)Using a neural network to approximate the state-action value function in reinforcement learning.The neural network replaces the Q table in reinforcement learning and solves the problem of ” curse of dimensionality ”.(2)Using the experience replay mechanism to ensure the quantity of neural network training data,and introducing heuristic knowledge to ensure the quality of neural network training data.Sufficient and effective training data speeds up the training,makes the algorithm converge faster,and improves the generalization of the model.(3)Applying the new method to the intelligent car and design a simulation experiment.When the intelligent car walks in an unknown environment,it can collect experience data in real time and use those data to train the neural network.The heuristic knowledge provides training data for the neural network while guiding the intelligent car selection action,which effectively helps the intelligent car avoid blind exploration.We performed simulation experiments on maps of different complexity to prove the effectiveness of our proposed method.