Transitioning Control and Sensing Technologies from Fully-autonomous Driving to Driver Assistance Systems
Based on our experience in the DARPA Urban Challenge and on current trends in consumer automobiles, we believe that driver assistance systems can be significantly improved by new techniques in control and sensing that have been developed for fully-autonomous driving.
In particular, from the control community, real-time Model Predictive Control (MPC) can be used as the next generation of cruise control for automobiles, offering a principled method for robustly incorporating information from automobiles' existing sensing systems, such as GPS and odometry, as well as from additional sensors that will be used in future, complimentary driver assistance systems, such as visible-light cameras, infrared (IR) cameras and laser scanners.
From the sensing community, we believe that obstacle-detection systems using passive (hence, noninterfering) and cost-effective visible-light cameras and thermal IR cameras, originally developed for fully-autonomous driving, are also a valuable addition to the driver assistance toolbox, offering the ability to warn drivers about moving or heat-producing obstacles, including pedestrians and other automobiles.
In this paper we will discuss methods, derived from fully-autonomous vehicle research, for real-time Model Predictive Control (MPC), segmentation of moving (relative to the ground) obstacles using visible-light cameras, and detection of heat-producing objects using thermal infrared (IR) cameras, as well as their application to driver assistance systems.