Deep Learning-Based Dynamic Path Planning and Optimization for Autonomous Vehicles

In the contemporary research context, path planning for autonomous vehicles in dynamic and uncertain environments represents a significant challenge. In this paper, we propose a dynamic path planning and optimisation framework for autonomous vehicles based on Deep Reinforcement Learning (DRL). This framework innovatively introduces real-time adaptability and an optimal path selection mechanism. The proposed model is based on DRL, with the addition of a convolutional neural network (CNN) and an attention mechanism to optimise the perception and decision-making process. In the context of path planning, CNNs facilitate the extraction of intricate features of the road environment, while the attention mechanism assists the model in prioritising critical environmental factors (e.g., traffic signals, obstacles, etc.), thereby enhancing the precision and efficiency of decision-making processes. Furthermore, the incorporation of reinforcement learning-based online learning capabilities enables the model to make adaptive adjustments according to real-time traffic conditions and dynamically plan the optimal path. Experimental results demonstrate that the proposed model exhibits superior performance in comparison to traditional path planning algorithms within a simulation environment, with an estimated enhancement in path planning efficiency of approximately 15%.