Decision Framework for Unsignalized Intersection Based on Discrete Soft Actor-Critic with Attention Mechanism and Occupancy Grid Integration

A novel decision framework for unsignalized intersection autonomous driving is proposed, targeting the optimization of safety and efficiency. The framework is based on the discrete soft actor-critic, specifically designed for discrete action spaces and occupancy grid inputs. By employing multi-dimensional occupancy grid inputs, the framework effectively circumvents data sequencing issues inherent to traditional kinematic data inputs. Channel and spatial attention mechanisms are integrated to focus on and process features, thereby enhancing the algorithm's performance. Experimental results demonstrate that compared to other algorithms, this framework achieves a better balance between safety and efficiency in autonomous driving. Moreover, it exhibits practicality and robustness in complex traffic scenarios, showcasing its potential superiority over existing methods.