No project description provided
Rex: an open-source domestic robot
This repository represent an experiment made using pyBullet and OpenAI Gym. It's a very work in progress project.
This project is mostly inspired by the incredible works done by Boston Dynamics.
The goal is to train a 3D printed legged robot using
Reinforcement Learning. The aim is to let the robot learns
domestic and generic tasks (like
pick objects and
autonomous navigation) in the simulations and then successfully
transfer the knowledge (Control policies) on the real robot without any other tuning.
Python 3.7 virtual environment, e.g. using
conda create -n rex python=3.7 anaconda conda activate rex
Install the public
pip install rex_gym
Install from source
You can also clone this repository and install it using
pip. From the root of the project:
pip install .
Run pre-trained agent simulation
To start a pre-trained agent:
There are also videos under
Start a new training simulation
To start a new training session:
python -m rex_gym.agents.scripts.train --config rex_reactive --logdir YOUR_LOG_DIR_PATH
YOUR_LOG_DIR_PATH sets where the policy output is stored.
PPO Agent configuration
You may want to edit the PPO agent's default configuration, especially the number of parallel agents launched in the simulation.
num_agents variable in the
def default(): """Default configuration for PPO.""" # General ... num_agents = 14
Install rex_gym from source. This configuration will launch 14 agents (threads) in parallel to train your model.
I've printed the components using a Creality Ender3 3D printer, with PLA and TPU+ (this last one just for the foot cover).
The idea is to extend the basic robot adding components like a 3 joints robotic arm on the top of the rack and a Lidar sensor.
Rex: simulation engine
Rex is a 12 joints robot with 3 motors (
Foot) for each leg.
pose signal (see
rex_reactive_env.py) sets the 12 motor angles that make Rex stands up.
The robot model was imported in
pyBullet creating an URDF file.
This is a very first list of tasks I'd like to teach to Rex:
- Locomotion - Run/Walk
- Stand up
- Reach a specific point
- Autonomous navigation - Map environment
- Grab an object
This task is about let Rex learns how to run in a open space.
Reinforcement Learning Algorithm
There is a good number of papers on quadrupeds locomotion, most of them comes with sample code. The most complete examples collection
is probably the Minitaur folder in the PyBullet3 repository.
This repository collects the code samples for the
I've extracted and edited the
Minitaur Reactive Environment, sample code for the paper Sim-to-Real: Learning Agile Locomotion For Quadruped Robots, and used it
to automate the learning process for the locomotion gait for Rex. I've tried to retain all the improvements introduced in that paper
to overcome the Reality Gap.
Galloping gait - from scratch
In this very first experiment, I let the system learn from scratch: I set the open loop component
a(t) = 0 and
gave the feedback component large output bounds
[−0.5,0.5] radians. The
leg model (see
forces legs and foots movements (positive or negative direction, depending on the leg) influencing the learning
score and time. In this first version, the
leg model holds the Shoulder motors in the start position (0 degrees).
As in the Minitaur example, I choose to use Proximal Policy Optimization (PPO).
I've ran a first simulation (~6M steps), the output
control policy is in
The emerged galloping gait shows the robot body tilled up and some unusual positions/movements (especially starting from the initial pose). The
leg model needs improvements.
The policy video is
Galloping gait - bounded feedback
To improve the gait, in this second simulation, I've worked on the
I set bounds for both
Foot angles, keeping the
Shoulder in the initial position.
I've ran the simulation (7M steps), the output
control policy is in
The emerged gait looks more clear. The policy video is
Sim-to-Real: Learning Agile Locomotion For Quadruped Robots and all the related papers. Google Brain, Google X, Google DeepMind - Minitaur Ghost Robotics.
Deok-yeon Kim creator of SpotMini.
The great work with the robot platform rendering done by Florian Wilk with his SpotMicroAI.
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size rex_gym-0.1.1.tar.gz (32.7 MB)||File type Source||Python version None||Upload date||Hashes View|