This is our final project for nonlinear control course. We are trying to implement different control method to control simple robotic biped walking.

The first controller was four independent PD controller to track reference joint signal. It was easy, but had bad performance when uncertainty existed.

The second controller was virtual model control from MIT, slope walking was improved.

The third controller was robust adaptive virtual model control, extra mass could be put on the robot and it still walks well.

I want to further investigate neural control of human walking and application to robot control. We tested a simple neural oscillator that can generate periodical signals for each joints to walk on plane.