Fault tolerant localization is a critical component of autonomous driving technology. In order to perform any control and perception of objects around its environment, a vehicle must possess an accurate estimate of its own location in the environment despite potential sensor failures. This paper describes the system and architecture developed to peform fault tolerant localization on a prototype test vehicle using a differential GPS and INS coupled system, a camera, and laser rangefinder sensors. A ROS-based system was developed to allow modularization and interfacing with other modules such as perception and control. The system permits parameter tuning and automated testing for both software-in-the-loop and hardware-in-the-loop, as well as interchanging of different algorithms for the localization itself. The framework for developing and testing new localization algorithms has been built in order to test a particle-filter based implementation that provides an accurate estimate of the vehicle's location despite failures in GPS, camera, or laser rangefinder.




Download Full History