Numerical derivatives of dynamics for mujoco

• Needs mujoco licence to run.
• Wraps derivative.cpp to call from Python.

Installation

1. Install mujoco_py
2. pip install mujoco_py_deriv

Usage

Prepare mujoco model.

import mujoco_py as mj
from mujoco_py_deriv import MjDerivative, checkderiv

# Prepare mujoco model and data
model = mj.load_model_from_path("flat_pusher_sample.xml")
sim = mj.MjSim(model, nsubsteps=nstep)
dmain = sim.data

Compute numerical derivative

# To compute δf/δx
f = ["qacc"]
x = ["qfrc_applied", "qvel", "qpos"]
deriv_obj = MjDerivative(model, dmain, f, x)
deriv = deriv_obj.compute()

Test case

python test/test_mujoco_py_deriv.py

Dynamics with derivatives

Take a look at mujoco_py_deriv_dynamics.py