Learning Coordinated Vehicle Maneuver Motion Primitives from Human Demonstration

摘要: High-fidelity computational human models provide a safe and cost-efficient method for studying driver experience in vehicle maneuvers validation of design. Compared to passive models, active capable reproducing the decision-making, as well maneuver motion planning control, will be able support realistic simulation human-vehicle interaction. In this paper, we propose an integrated interaction framework which learns primitives from drivers, uses them compose natural contextual driving motions. Specifically, recruited six experienced drivers recorded their motions on fixed-base testbed. We further segmented classified collected data based similarity joint coordination. Using combination imitation learning methods, extracted regularity variability across subjects, learned dynamic used reproduction simulation. present implementation lower-extremity coordination pedal activation longitudinal control. Our research efforts lead primitive library enables motions, with dynamics proposed simulating