Go2 Convex MPC — Glove Teleoperation

Real-time teleoperation of the Unitree Go2 quadruped via a custom sensor glove, built on top of a Convex MPC locomotion controller in MuJoCo simulation.

This project was developed as part of the Yonsei University Human-Centered AI Robotics Lab's 2025 Winter Internship Program, and won 1st place at the Yonsei University VAR Program Final Symposium.

---

Original Repository

This repo is based on go2-convex-mpc by elijah-waichong-chan. The original work implements a Convex MPC locomotion controller for the Unitree Go2 quadruped in MuJoCo, developed as a UC Berkeley MEng capstone project.

---

What's Added

• Glove teleoperation (teleop/) — real-time robot control via a custom Arduino glove with a magnetometer, haptic motor, and capacitive finger sensors
• Finger combo → feature mapping (stand/sit, locomotion, recovery, Euler lean)
• Demo runner for quick simulation start

---

Installation

git clone https://github.com/elijah-waichong-chan/go2-convex-mpc.git
cd go2-convex-mpc
conda env create -f environment.yml
conda activate go2-convex-mpc
pip install -e .

---

Run

Glove teleoperation:

python teleop/run_teleop.py --port /dev/ttyACM0

Quest3 Remote controller (requires ALVR):

python teleop/run_quest3.py

Experiment mode:

python teleop/run_experiment.py

Demo With HUD overlay:

python teleop/run_demo.py

MPC examples (from original repo):

python -m examples.ex00_demo
python -m examples.ex02_trot_forward