A simple box robot, eight joints learn to walk forward, for the observation state, there are X-axis angle, Y-axis angle, angle value of each leg. Incentive strategy: X-axis displacement difference *50. Behavior: The angle at which each leg will turn next time, the speed is based on the angle difference! (My strategy here may have problems, in perfection) The Z-axis displacement is fixed and only the forward speed is trained.
For more information:
My blog