Crowd Navigation-Based on Attention Mechanism and Game Theory

This paper tackles the problem of ensuring provably safe and computationally efficient local obstacle avoidance for robots navigating dynamic and spatially constrained settings. Traditional reinforcement learning algorithms often overlook the heterogeneity and interaction of human behaviors, limiting their effectiveness in real-world settings. To address this, we propose a neural network that integrates graph and attention mechanisms, enhancing the robot’s ability to interpret pedestrian characteristics and intentions. Additionally, we incorporate game theory into the reinforcement learning reward function, guiding the robot to learn more effective human-robot interaction strategies. Experimental results demonstrate that our approach enables robots to perform well in complex crowd navigation scenarios, adeptly handling pedestrians with varying goals and avoidance intentions.