Online Terrain Classification for Team CalTech in DARPA Grand Challenge

Abstract: The US Department of Defense wants one third of ground military forces to be automated by 2015. By using self-driving military vehicles, American troops can be kept out of harm's way in the battlefield.1 To speed up the development, the US Department of Defense decided to arrange a desert race for autonomous vehicles, DARPA Grand Challenge. Team Caltech is entering the Challenge with the vehicle Alice. She is equipped with sensors that scan the surroundings and a GPS that gives Alice her location. The GPS signal is the only type of communication that is allowed during the race. This master's thesis describes the work that Team Caltech did to prepare for DGC. The thesis also describes the work that I did as a member of Team Caltech: administration of the testing and on-line terrain classification. Testing is a critical part in the development of an autonomous vehicle. As time is limited, it is essential to test the right things and the right types of terrain. To improve the efficiency and effectiveness of the testing, I developed a test matrix for different terrain types. This matrix helped us to make sure that we were testing different types of terrain. Having developed the matrix, I was also responsible for the documentation of the testing. The vehicle acts differently depending on the type of terrain. Thus, it is valuable that Alice can differentiate the current terrain type. I have written a program in C++ that can distinguish and classify different terrain types. The C++ program was not implemented and tested in real conditions, because time was too short. However, it will be useful for the future.