Abstract

The goal of this project is to efficiently produce robust locomotion for a commercial quadrupedal robot platform Unitree A1 by using a model-based controller, more specifically a model predictive controller. The main contribution of this work is twofold: first, we set up the framework for controlling A1 based on the Robot Operating System. The robot's states are updated in our simulation platform. Second, we applied a model predictive control approach for the A1 robot model. The model predictive control approach can perform more robust locomotion tasks as it finds an optimal system input to track command trajectories in a long time horizon while taking into account the system's dynamics. Finally, we proposed a more computationally efficient MPC formulation by effectively reducing the number of optimization variables and constraints that results in real-time execution on A1 robot's onboard PC with limited computational power. Our formulation is 6 times faster than the previous MPC formulation. With our formulation, the MPC-based controller can be run in real-time with a time horizon of 1.2 seconds on A1, which is impossible for the previous formulation.

Video of synchronization

Simulation of the new efficient Model Predictive Controller

Simulation of the old slow Model Predictive Controller

Acknowledgements

The website template was borrowed from Michaël Gharbi.