ANCP: Attention-Based Neural Circuit Policies for Robot Spatiotemporal Reasoning in Crowd Navigation

This paper explores the complexities of autonomous navigation for mobile robots in densely populated environments, where the accuracy of crowd behavior modeling is crucial. Recent studies have increasingly leveraged deep reinforcement learning and biologically inspired methodologies, which demonstrate enhanced learning capabilities and adaptability in intricate settings. This work proposes a novel attention-based neural circuit policies (ANCP) that merges biological neural mechanisms with attention processes to facilitate spatiotemporal reasoning in robot navigation. We evaluated the algorithm in demanding simulation settings, and the findings indicate that our approach adapts more effectively to crowded scenarios compared with the baseline algorithms.