BiConMP MuJoCo

MuJoCo wrapper for BiConMP solver in a docker container.

---

Install

This project uses VS Code devcontainer. To run this workspace, please first install VS Code with the following extensions:

• Dev Containers (ms-vscode-remote.remote-containers)
• Remote Developement (ms-vscode-remote.vscode-remote-extensionpack)
• Docker (ms-azuretools.vscode-docker)

After that is done, clone this repository onto your developement PC.

git clone --recurse-submodules https://github.com/Atarilab/biconmp_mujoco.git

Then, enter the directory downloaded and open it in vscode

cd biconmp_mujoco && code .

After VS Code has started, there should be a prompt to start this workspace as a container. Otherwise, you can also do ctrl + shift + p then select Dev Container: Rebuild and Reopen in Container to start it manually.

The environment is set up automatically with BiConMP installed.

Wrapper interface

This repo provides a wrapper to interface MuJoCo for the simulation environment and Pinocchio used by BiConMP. The source code is in project/example/mj_pin_wrapper.

The wrapper is imported as a submodule in the project from this repository.

Pinocchio + MuJoCo interface

Defined in the RobotWrapperAbstract class. This class provides tools to interface model descriptions from both MuJoCo and Pinocchio. It also checks that both descriptions are similar.

Paths of urdf and mjcf (as xml) description files needs to be provided. One can use RobotModelLoader to get the paths automatically given the robot name (now only working with go2 robot).

• get_pin_state and get_mj_state return the state of robot respectively in Pinocchio and MuJoCo format (different quaternion representation). One can also access to the joint names or end effector names.

• get_mj_eeff_contact_with_floor returns which end effectors is in contact with the floor in the simulation as a map {end effector name : True/False}

QuadrupedWrapperAbstract inherits from RobotWrapperAbstract and provide additional tools for quadruped robots by abstracting the robot description.

• get_pin_feet_position_world returns the feet positions in world frame (same for hip, thigh and calf).

Simulator

For custom usage of the wrapper, two abstract classes used in the Simulator need to be inherited.

1. ControllerAbstract Implements the robot controller or policy. The following method should be inherited.

   • get_torques(q, v, robot_data) This method is called every simulation step by the simulator. It takes as input the robot position state q: np.array, the robot velocity state v: np.array and the simulation data robot_data: MjData. It should return the torques for each actuator as a map {joint name : $\tau$}. One can find the joint names the RobotWrapperAbstract.

   See BiConMPC for exemple of inheritance.

2. DataRecorderAbstract Implements a class to record the data from the simulation. The following method should be inherited.

   • record(q, v, robot_data) This method is called every simulation step by the simulator. It takes as input the robot position state q: np.array, the robot velocity state v: np.array and the simulation data robot_data: MjData.

One can run the simulation with or without viewer using the run method.

Usage

See main.py for a minimal example.