Immersive Traffic Interactive Simulator for Multi-Agent

Interaction in complex traffic scenarios poses significant risks, but existing research lacks a thorough analysis of interaction mechanisms. This gap presents challenges for self-driving cars to navigate such complex scenarios safely. Mining interaction mechanisms heavily rely on real-world data, which is costly and risky. Therefore, interaction data acquisition often occurs through simulation test environments. However, current driving scene simulators typically use predefined rules to simulate interactions, failing to represent genuine responses in real scenarios accurately. This study proposes an immersive interaction simulator system for multiple agents to address this limitation and obtain authentic interaction data within simulations. The simulator covers five common interaction cases in traffic scenarios, and three interaction types: vehicle-vehicle, human-vehicle, and pedestrian-vehicle-vehicle interactions. Vehicle interaction data is incorporated using driving simulator devices, while pedestrian interaction data is captured through virtual reality (VR) and motion capture devices. Real-time closed-loop feedback of simulated interactions is ensured by implementing a multi-computer communication protocol. Furthermore, the system includes a comprehensive data recording module, enabling playback analysis and rewriting of interactive cases. The development of this interaction simulator holds promise for deepening our understanding of interaction mechanisms and has potential applications in autonomous driving, thereby enhancing the safety of self-driving cars.