Research On Driving Behavior Decision-making Of Real-world Link Robot Based On Reinforcement Learning

With the development of artificial intelligence,more and more intelligent applications are changing our way of life.Unmanned driving in the automotive field is an important research direction for major car companies in the future.The research on the decision-making behavior of robots based on reinforcement learning in the real world is very important to the field of unmanned driving,and has very broad application prospects and research significance.Therefore,this paper proposes to independently build a training environment close to the real-world road conditions,which can provide a training real-world road environment for the reinforcement learning algorithm.Integrate the road types,weather conditions,congestion conditions,and road speed limit requirements of real-world roads into the built environment.Based on the reinforcement learning algorithm,the driving decision-making actions that conform to the general driving rules are set in this environment,and the corresponding reward values are set reasonably.Combining the influence of the real world on driving decision-making and the situation of road information generation for conversion design.Identify the real-time congestion situation of roads in the real world,and update the congestion situation of each small section of the road that the agent travels to.Design the agent to effectively determine the congestion method and take corresponding deceleration measures.Improve the intelligent body to reach the end point as soon as possible under the premise of ensuring no collision.This paper uses the DDQN algorithm to conduct simulation training and verification of the stability and fluency of the environment,and uses DQN and Dueling DDQN to conduct comparative experiments.It is proved that the construction of this environment satisfies the research on driving behavior decision-making based on the reinforcement learning algorithm,and the rationality of each interaction process in the environment is analyzed through the simulation results.As the number of training rounds increases,the obtained reward value also generally tends to rise,and it can be concluded that the effectiveness of driving behavior decision-making based on reinforcement learning.