Reliable Vulnerable Road Users Behaviour Prediction Considering Spatiotemporal and Socialized Interactions

ongoing

As a connection module between the perception layer and decision-making control layer of autonomous vehicles, behaviour prediction is one of the research focuses in this field. The existing behaviour prediction technoAs a connection module between the perception layer and decision-making control layer of autonomous vehicles, behaviour prediction is one of the research focuses in this field. The existing behaviour prediction technologies have the problems of unclear spatiotemporal interaction coupling mechanism and socialized interaction mechanism between the target object and environment, which makes it difficult to achieve accurate behaviour prediction and seriously restricts the practical application of this technology in the field of autonomous vehicles. Aiming at complex high-load mixed traffic flow scenarios, this project takes Vulnerable Road Users (VRUs, e.g. pedestrians, cyclists, etc.) as the research objects, and conducts research on the theories and key technologies of behaviour prediction driven by spatiotemporal and socialized interactions. The content includes: VRUs-vehicle coupling mechanism investigation in spatiotemporal domain and spatiotemporal interaction modelling; VRUs-vehicle socialized mechanism study in social psychology aspect and adaptive socialized interaction modelling; synergistic driven mechanism study in multi-model integration field and dynamic ensemble learning modelling; and validation and optimization of VRUs behaviour prediction models. The vision of ReSIN is to develop an accurate, interpretable and robust/resilient VRUs behaviour prediction model for AVs to behave safely, efficiently and confidently under complex and uncertain traffic environments. By empowering AVs with this prediction capability, ReSIN will enable a seamless integration into the decision-making and planning modules of AVs, which will allow AVs to comprehend and predict the intricate patterns and variability in VRU behaviour in dynamic changing environments.