Pedestrian-vehicle Interaction and Autonomous Vehicles
Introduction
This research focuses on the critical challenge of human-centric autonomous driving, specifically modeling the complex, social interactions between vehicles and pedestrians. By integrating multi-agent 3D skeletal motion modeling with decision-making frameworks, the work seeks to make automated vehicles (AVs) more predictable and "socially aware". A key contribution of this research is the introduction of Social Value Orientation (SVO) into dynamical models [Crosato et al. IEEE TIV2022, Crosato et al. IEEE ICHMS2021], allowing AVs to navigate interactive environments with a nuanced understanding of human cooperation and intent. This transition from reactive obstacle avoidance to interaction-aware decision-making is further supported by large-scale datasets like Waymo-3DSkelMo [Zhu et al. ACM MM2025], which provide the high-fidelity skeletal data necessary for vehicle-conditioned pose forecasting [Zhu et al. IEEE ICRA2026].
To ground these theoretical models in reality, the research also develops innovative Virtual Reality (VR) frameworks [Crosato et al. HRI2024] for safe and cost-effective data collection. These platforms allow for the study of human-driver interactions in high-risk scenarios without physical danger, bridging the gap between computer vision - such as multi-task deep learning using optical flow [Hu et al. IET IES2020] - and real-world robotics. From winning Best Paper Awards for pedestrian interaction [Crosato et al. IEEE ICHMS2021] to featuring in the top 10% of most-viewed intelligent systems research [Crosato et al. AIS2023], this body of work establishes a comprehensive pipeline for creating AVs that are not just technically proficient, but socially integrated into our urban landscapes.
To ground these theoretical models in reality, the research also develops innovative Virtual Reality (VR) frameworks [Crosato et al. HRI2024] for safe and cost-effective data collection. These platforms allow for the study of human-driver interactions in high-risk scenarios without physical danger, bridging the gap between computer vision - such as multi-task deep learning using optical flow [Hu et al. IET IES2020] - and real-world robotics. From winning Best Paper Awards for pedestrian interaction [Crosato et al. IEEE ICHMS2021] to featuring in the top 10% of most-viewed intelligent systems research [Crosato et al. AIS2023], this body of work establishes a comprehensive pipeline for creating AVs that are not just technically proficient, but socially integrated into our urban landscapes.
Publications
The Team