The Cassie Robot Research

I worked on this project in collaboration with Dr. Fares Abu-Dakka to develop an advanced control system for bipedal locomotion robots and demonstrate it on the Cassie Robot. A Robust high order non-linear controller is integrated with Hybrid Zero Dynamics (HZD) controller as well as a Neural Network (NN) to control the robot. The incorporation of HZD has a crucial role in reducing the size of the NN; moreover, the robustness of the robot's locomotion is improved with the addition of the robust non-linear controller. Proximal Policy Optimization is used to train the NN on various velocities.

Following are some of the preliminary results:

In the video above, the high-order non-linear controller is applied to control the  foot placement in longitudinal axis (x-axis) to stabilize the velocity control in the axis; whereas HZD is applied to control the foot placement in the lateral axis (y-axis), to control the torso orientation, and to control the trajectory tracking. As seen in the video the robot manages to fast track the changing desired velocity signal in the sagittal plane. Moreover, it is capable to fast recover after the application of the forces. 

In Fig. 1  and Fig. 2, the robot manages to stably track longitudinal velocities that range from -0.3 to 0.8 (m/s). Fig. 3 and Fig. 4 shows how the robot is able to fast recover after the force is applied to the torso in different configurations.