go1-MPC

MuJoCo MPC (MJPC) is an interactive application and software framework for real-time predictive control with MuJoCo, developed by DeepMind. MJPC allows the user to easily author and solve complex robotics tasks, and currently supports three shooting-based planners: derivative-based iLQG and Gradient Descent, and a simple yet very competitive derivative-free method called Predictive Sampling. https://github.com/deepmind/mujoco_mpc (For detailed usage of MuJoCo MPC refer this link)

The go1-MPC repo is based on MuJoCo MPC and additionally includes Unitree 0o1 robot simulation.

Compiler setup

Build in VSCode. We used GCC 11 compiler.

Required Changes to be done for Data Collection

Navigate to build/_deps/mujoco-src/src/engine/engine_print.c in VSCode. In mj_printFormattedData, line 744, change file mode from "wt" (write) to "a" (append).

Usage of go1-MPC

To run the simulation;

cd mujoco_mpc_v2_DC/mjpc/build
./mjpc

or use cmake tool to debug from VSCode while will run the target file(mjpc).

GUI usage

Below mentioned are a few key features.

• File has the options to print model and print data.
• In Options enable the Right UI to see the joint and control data.