Research On Intelligent Vehicle Dynamic Path Planning Algorithm Based On Improved Q-Learning

The development of intelligent vehicle technology not only improves the vehicle's own safety performance,but also effectively avoids the occurrence of traffic accidents and protects people's lives and property.The path planning technology has always been the core content of the development of intelligent vehicle technology,and it is the demand of social development.Reinforcement learning is an online learning mode in which Agent and environment continuously interact.Q-learning is an important reinforcement learning algorithm,which is very suitable for intelligent vehicle to carry out path planning in unknown environment.However,in the face of complex environmental changes,the algorithm is inefficient in learning and slow in convergence.Ant colony algorithm is a kind of swarm intelligence optimization algorithm.The algorithm can quickly find the local optimal solution,has strong robustness and adaptability,but the convergence speed is slower in the optimization process,and it is easy to fall into the local optimum.For the problem of intelligent vehicle path planning,this thesis firstly improves the traditional ant colony algorithm,formulates the pheromone local updating rules,and replaces the constant B of the traditional ant colony algorithm with the dynamic function A,which makes the ant optimize the current path in the optimization process.The search is more oriented,prompting ants to search for paths that have not been taken and avoid the algorithm being trapped in a local optimal solution.Secondly,this paper proposes an algorithm based on improved ant colony and Q-learning.By integrating the pheromone?^?_ij in the ant colony algorithm into the Q-learning,the intelligent vehicle can make full use of the combined effect of Q value and pheromone?^?_ij behavior decision and improve the learning efficiency.In the environment of MATLAB,the intelligent vehicle path planning environment model and the visual simulation interface were established.The simulation analysis and comparison were carried out in different static environments respectively.It verified the effectiveness of the improved ant colony algorithm combined with Q-learning in static path planning.Finally,this paper proposes a dynamic path planning method based on fuzzy control combined improved ant colony algorithm and Q-learning.A fuzzy inference system was established to control the pheromone local renewal volatilization factor?in real time,reducing the pheromone volatilization on the routes with better paths and less obstacles around the state of the intelligent vehicle.And by monitoring the changes in the number of reachable nodes in a certain range,the occurrence and location of dynamic obstacles are judged,and a path to bypass the obstacles to reach the target point is re-planned.