My university had a program that did this, I took it as a sophomore years ago while I was studying "evolutionary robotics", although typically we would simulate the robot in 3D virtual environment to test the reward mechanism for learning to walk.