Model the actuators, set up the simulators, train a walking policy with reinforcement learning, and transfer it to the real robot.