The Learning Locomotion project is a DARPA progam with the goal of developing a new generation of learning algorithms to enable traversal of rough terrain by unmanned ground vehicles (UGV). Typically, UGVs are not capable of traversing rough terrain, limiting their available area of operation. This project looks to develop new control algorithms that employ advanced learning techiniques in order to advance UGV capabilities in rough terrain.
The project has six research teams that both collaborate and compete. Each team is provided with the same small quadruped robot (Little Dog), built by Boston Dynamics. Each team is also provided with an advanced motion capture system by Vicon in order to reduce the complexity of sensing in the navigation problem. By using the same hardware, the research teams can focus on the problem of developing effective algorithms for negotiating rough terrain.