RBE PhD Qualifier Presentation
Computational Human Modeling for Human-Vehicle Interaction
Kenechukwu C. Mbanisi
Abstract: Recent advances in the development of active vehicle safety and intelligent driving systems are owed partly to the use of high-fidelity computational driver models. These models enable predictive simulation and analysis of driver behavior in a broad range of driving scenarios. However, existing models paid limited attention to realistic rendering of maneuver motions employed in steering and pedal activation tasks.
In this presentation, I will describe our proposed systematic approach for learning the motion primitives of vehicle maneuver motions from human drivers, and use them to compose natural and contextual driving motions in simulation. Specifically, we recruited eight experienced drivers and recorded their vehicle maneuver motions on a fixed-base driving simulation test bed. We further extracted a set of characteristic vehicle maneuver motion styles from the demonstration data. Using a combination of imitation learning methods, we extracted the regularity and variability of vehicle maneuver motion styles across participants, and modeled them as motion primitives that can be used for motion reproduction in simulation.
Our proposed method enables the learning of coordinated steering and pedal activation motions from human driving data, and the reproduction of whole-body coordination on our integrated cognitive-physical simulation framework for human-vehicle interaction.
Professor Mike Gennert
Professor Jie Fu
Professor Zhi Li (Advisor)