Modeling Vehicle-Type-Specific Pedestrian Crash Avoidance Behavior in Safety-Critical Interactions Using Smooth-Mamba Deep Reinforcement Learning
Researchers developed SMamba-DDPG, a deep reinforcement learning framework that models how pedestrians behave differently when interacting with autonomous vehicles versus human-driven vehicles. The study found that pedestrians react faster to AVs and adopt lower crossing speeds, with AV interactions showing lower conflict rates than HDV scenarios.
This research addresses a critical gap in autonomous vehicle safety by quantifying behavioral differences in pedestrian-vehicle interactions across vehicle types. Using real-world crash avoidance data from the Argoverse 2 dataset, the team developed a sophisticated model that captures pedestrian decision-making under safety-critical conditions. The findings reveal that pedestrians exhibit measurably different behavioral patterns depending on whether they're interacting with autonomous or human-driven vehicles, suggesting that current AV safety systems may benefit from pedestrian-behavior-aware design principles.
The research reflects growing recognition that AV deployment cannot rely solely on vehicle-centric safety models. As mixed-traffic environments become reality, understanding how vulnerable road users adapt their behavior to different vehicle types becomes operationally critical. The discovery that pedestrians respond faster to AVs and use lower crossing speeds indicates that perception of automation influences human decision-making in safety-critical moments.
The commercial implications extend across multiple domains. AV manufacturers gain data-driven insights for tuning vehicle behavior to improve pedestrian interactions. Simulation companies can integrate vehicle-type-specific pedestrian models into traffic modeling tools, improving the fidelity of autonomous driving validation environments. Insurance and safety organizations can use these behavioral insights to assess real-world safety outcomes more accurately.
Moving forward, validation of these findings across diverse pedestrian populations and geographic contexts remains essential. The framework's applicability to other safety-critical scenarios—such as urban intersections with varying infrastructure—warrants investigation. Integration of these behavioral models into AV development pipelines could represent a meaningful safety improvement.
- →Pedestrians exhibit demonstrably different crash avoidance behaviors when interacting with autonomous versus human-driven vehicles
- →SMamba-DDPG framework successfully captures vehicle-type-specific behavioral patterns with higher accuracy than baseline models
- →Pedestrians respond faster and maintain lower crossing speeds with AVs, indicating faster threat perception
- →Pedestrian-AV interactions show lower conflict rates and higher yielding rates compared to pedestrian-HDV interactions
- →Vehicle-type-specific behavioral models are essential for realistic mixed-traffic simulation and safer AV system design