Bipedal Hopping: Reduced-order Model Embedding via Optimization-based Control

Xiaobin Xiong,Aaron Ames

DOI: https://doi.org/10.48550/arXiv.1807.08037

2018-08-03

Abstract:This paper presents the design and validation of controlling hopping on the 3D bipedal robot Cassie. A spring-mass model is identified from the kinematics and compliance of the robot. The spring stiffness and damping are encapsulated by the leg length, thus actuating the leg length can create and control hopping behaviors. Trajectory optimization via direct collocation is performed on the spring-mass model to plan jumping and landing motions. The leg length trajectories are utilized as desired outputs to synthesize a control Lyapunov function based quadratic program (CLF-QP). Centroidal angular momentum, taking as an addition output in the CLF-QP, is also stabilized in the jumping phase to prevent whole body rotation in the underactuated flight phase. The solution to the CLF-QP is a nonlinear feedback control law that achieves dynamic jumping behaviors on bipedal robots with compliance. The framework presented in this paper is verified experimentally on the bipedal robot Cassie.

Robotics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to achieve controllable jumping behavior on the 3D bipedal robot Cassie. Specifically, the researchers synthesize a nonlinear controller by identifying the spring - mass model of the robot and embedding this model into the full - dynamic model. The core objective of the paper is to design a control strategy so that the bipedal robot with mechanical compliance can perform dynamic jumping actions while preventing the rotation of the entire body during the flight phase. To achieve this goal, the researchers adopt the trajectory optimization method to plan the jumping and landing motions, and use the quadratic programming based on the control Lyapunov function (CLF - QP) to stabilize the angular momentum of the center of mass, thereby ensuring the stability and controllability of the robot during the jumping process.

The key technical points of the paper include:
1. **Identification of the spring - mass model**: Starting from the dynamic characteristics and compliance of the robot, a spring - mass model is identified, which can simulate the influence of the leg - length change of the robot on the jumping behavior.
2. **Trajectory optimization**: The trajectory of the spring - mass model is optimized by the direct collocation method to generate the motion trajectories of jumping and landing.
3. **Quadratic programming based on the control Lyapunov function (CLF - QP)**: The CLF - QP is used to synthesize the controller to ensure that the robot can keep the angular momentum of the center of mass stable while tracking the predetermined leg - length trajectory, thus avoiding the overall rotation during the un - actuated flight phase.
4. **Experimental verification**: Experimental verification is carried out on the 3D bipedal robot Cassie, demonstrating the effectiveness of this control method.

Through these technical means, the researchers have successfully achieved the jumping behavior of Cassie, reaching a take - off height of about 7 inches and an air - time of 0.423 seconds, and the ground reaction force and take - off time of the robot are highly consistent with the prediction results of the spring - mass model, verifying the reliability and effectiveness of the proposed method.