Location: Ann Arbor, MI | 09/2025 - 02/2026

Github: https://github.com/ruben-fonseca-castro/WBIC_MPC_Project

 

This project involved the implementation and experimental tuning of a hierarchical whole-body controller for quadruped locomotion based on convex Model Predictive Control (MPC) and Whole-Body Impulse Control (WBIC) using Null Space Projection for task prioritization (https://arxiv.org/abs/1909.06586). The paper itself originally implemented the controller on the MIT Mini-Cheetah.

 

For our implementation, the controller was instead implemented on a Unitree A1 quadruped robot in MuJoCo and achieved 40 seconds of forward walking, traveling 2.0 m. Through systematic experimentation, we identified critical control challenges including persistent oscillatory behavior (±10-15° in roll/pitch), velocity undershoot (33\% of commanded), and front-rear leg tracking asymmetry. We document the complete implementation with detailed mathematical formulations, experimental results with quantitative analysis, and insights into the limitations of linearized MPC at larger orientation angles.

​

This project is ongoing as I work to tune and improve the performance of both quadratic programs to achieve stability akin to currently deployed quadruped systems.