Agressive Autonomous driving


Driving at high speeds in unknown, offroad environments (such as those encountered in a rally race) offers a great number of challenges to control and vision researchers. Vision tasks become challenging as speed increases due to reduced processing time between seeing an object and needing to react to it. In unknown and semi-structured environments, different cues must be processed in order to find salient features such as road boundaries. The control tasks are equally challenging, where controllers need to deal with highly nonlinear systems with strong external disturbances (such as potholes and surface changes) and changing and difficult to estimate surface friction.

In order to research these challenges, a platform is needed that is high performance, reliable, and has sufficient computing power. The AutoRally platform meets these objectives. It is based on off-the-shelf parts, including an HPI baja 5sc 1:5 scale RC car chasis and Ubuntu/ROS based computer with a high performance GPU. All of these parts are selected for ease of use and robustness. The integrated system is designed to withstand crashes and roll-overs with minimal damage while still affording high performance and ease of testing.


Aggressive Driving with Model Predictive Path Integral Control

Grady Williams, Paul Drews, Brian Goldfain, James M. Rehg, and Evangelos Theodorou. “Aggressive Driving with Model Predictive Path Integral Control.” Accepted for publication in IEEE Intl. Conf. on Robotics and Automation (ICRA ’16)