gmargo11 / Cheetah-Software

Cheetah-Software: Gabe's Fork

This repository is forked from mit-biomimetics/Cheetah-Software.

Building

cd Cheetah-Software
mkdir mc-build && cd mc-build
cmake -DPYTHON_EXECUTABLE:FILEPATH=/usr/bin/python3.6 ..
make -j4
make install

Python Binding Usage

ldconfig -v
export PYTHONPATH=$PYTHONPATH:$LOCAL_BUILD/lib

Then, in python:

import pycheetah

Testing the Bindings

There are simple tests for the python bindings in python/binding_tests.py.

Interfacing the Controller with RaiSim

You will need to install RaiSim and the Raisimpy python bindings (https://github.com/robotlearn/raisimpy).

Then, run the python script python/raisim_control.py which executes the MIT Controller on the mini cheetah in RaiSim. The video is saved as video/raisim.mp4.

Here's a video of mini-cheetah trotting on rough terrain in RaiSim using Convex MPC and WBIC.

If you are running on a remote server, you may want to use xvfb to generate video headlessly: xvfb-run -a --server-args="-screen 0, 1280x1024x24" python3 raisim_control.py

In RaiSim, you can easily generate custom terrains, simulate objects, adjust surface properties, and more.

Adjusting the Gait Parameters

The NeuralGaitMPC controller is under construction. Currently, it accepts the gait parameters (offset, duration, velocity) as an argument to its run function. You can see an example usage of the python bindings in python/raisim_control_adaptive_gait.py.

Here's a video of mini-cheetah making a preprogrammed gait switch on rough terrain.