Research on Dynamic Path Planning for Autonomous Vehicles Based on Reinforcement Learning

The ability to plan routes in a dynamic manner is crucial for ensuring the safe and efficient operation of autonomous vehicles in complex and dynamic environments. This paper proposes an enhanced Deep Q Network (DQN) model for the realtime optimisation of dynamic paths. The proposed model employs the Deep Q-Network to process continuous sensor data, thereby enabling the mapping of environmental states and optimal actions for obstacle avoidance and route selection. In contrast to conventional methodologies that depend on fixed cartographic representations and predefined itineraries, our approach enables the vehicle to learn through interaction with the environment, thereby facilitating adaptation to evolving circumstances and uncertainties. An innovative adaptive reward function was designed to balance safety, efficiency, and passenger comfort, adjusting to real-time traffic conditions and environmental changes. This encourages the vehicle to make decisions that optimise overall performance. The experimental results demonstrate that our model exhibits a notable superiority over existing path planning algorithms in terms of collision avoidance, reduction in driving time, and adaptation to dynamic scenarios in simulated urban driving environments.