quadruped_robot

MIT mini cheetah simulation in pybullet

MIT mini cheetah use customized simulator and lcm framework, which is not a popular way to do the robot development. Now, we extract the algorithm and do the simulation using ros and pybullet. This can be simple to deploy the system into different custom robot or plantform, and easy to learn the algorithm.

System requirements:

Ubuntu 20.04, ROS Noetic

Dependency:

Clone all three repos under same directory with this repo.

# use Logitech gamepad to control robot
git clone https://github.com/Derek-TH-Wang/gamepad_ctrl.git

# msg rospack and rviz plugin
git clone https://github.com/loco-3d/whole_body_state_msgs.git
git clone https://github.com/eborghi10/whole_body_state_rviz_plugin.git

Run

We run this inside a docker container.

Start docker

Build docker image and start container.

cd docker
docker build -t ros-desktop .

xhost +
./run_docker.sh

Build

docker attach

cd {your workspace}
catkin make
source devel/setup.bash

Install Python dependencies

pip3 install -r requirements.txt

Terrain

you can modify the config/quadruped_ctrl_cinfig.yaml/terrain to deploy different terrains, there are four terrains supported in the simulator now, for example:

"plane"
"stairs"
"random1"
"random2"
"racetrack"

Running:

run the gamepad node to control robot:

roslaunch gamepad_ctrl gamepad_ctrl.launch

run the controller in simulator:

roslaunch quadruped_ctrl quadruped_ctrl.launch

switch the camera on / off: camera set True or False in config/quadruped_ctrl_config.yaml, then launch the rviz to see the point cloud:

roslaunch quadruped_ctrl vision.launch

also can switch the gait type:

rosservice call /gait_type "cmd: 1"

gait type:

0:trot
1:bunding
2:pronking
3:random
4:standing
5:trotRunning
6:random2
7:galloping
8:pacing
9:trot (same as 0)
10:walking
11:walking2

qZhang88/quadruped_ctrl