2020/06/01 by Cyrus Anderson, Ram Vasudevan, Anderson, Cyrus +3
Engineering · #Autonomous Vehicle Technology and Safety #FOS: Computer and information sciences #Robotics (cs.RO) #Traffic Prediction and Management Techniques #Traffic and Road Safety
paper · pdf · doi:10.48550/arxiv.2006.00962
openalex publication_date 2020/06/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Pedestrians and drivers interact closely in a wide range of environments.\nAutonomous vehicles (AVs) correspondingly face the need to predict pedestrians'\nfuture trajectories in these same environments. Traditional model-based\nprediction methods have been limited to making predictions in highly structured\nscenes with signalized intersections, marked crosswalks, or curbs. Deep\nlearning methods have instead leveraged datasets to learn predictive features\nthat generalize across scenes, at the cost of model interpretability. This\npaper aims to achieve both widely applicable and interpretable predictions by\nproposing a risk-based attention mechanism to learn when pedestrians yield, and\na model of vehicle influence to learn how yielding affects motion. A novel\nprobabilistic method, Off the Sidewalk Predictions (OSP), uses these to achieve\naccurate predictions in both shared spaces and traditional scenes. Experiments\non urban datasets demonstrate that the realtime method achieves\nstate-of-the-art performance.\n