Stanford's Autonomous Cars Take To The Racetrack

Stanford University researchers have been working on developing autonomous cars to make them safer and better able to cope with unknown conditions. To do this, they have taken the Stanford autonomous VW GTI, known as Niki, and the autonomous Audi TTS, known as Shelley, to Thunderhill Raceway Park. The goal was to help the autonomous autos to perform safely in extreme and unknown circumstances.Stanford says that the autonomous systems in both cars performed "about as well as an existing autonomous controls system" the team says that the system also performed as well as an experienced racecar driver. Stanford as a stated goal of making its autonomous control systems as good or better than the best, skilled drivers.

The systems can learn from past driving maneuvers. The team says that control systems for autonomous cars need access to information about road-tire friction. That info dictates limits of how hard the car can brake, accelerate, or steer and still stay on the road during emergency maneuvers. The challenge is that friction in the real world is difficult to predict and variable.

The team has developed a neural network integrating data from past driving experiences at Thunderhill along with information from a winter driving test facility to provide foundation knowledge consisting of 200,000 physics-based trajectories. The team is trying to blend data-driven methods and approaches that are grounded in fundamental physics to utilize the strengths of the individual techniques.

In testing, Shelley went around the track controlled by the physics-based autonomous system pre-loaded with data about the course and conditions. The team says that using that data, Shelley and a skilled amateur driver generated comparable lap times. Niki drove using the new neural network and performed similarly to the physics-based system. This is despite the fact that Niki lacked explicit information about road friction. The team has stressed that the neural network doesn't perform well in conditions outside those it has experienced.