rex-gym 0.1.2

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.

Installation

Create a Python 3.7 virtual environment, e.g. using Anaconda

conda create -n rex python=3.7 anaconda
conda activate rex

PyPI package

Install the public rex-gym library:

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:

python rex_gym.playground.rex_reactive_env_play

There are also videos under /videos.

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. Edit the num_agents variable in the agents/scripts/configs.py script:

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.

Robot platform

The robot used for this experiment is the Spotmicro made by Deok-yeon Kim.

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 (Shoulder, Leg and Foot) for each leg. The Rex 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.

Tasks

This is a very first list of tasks I'd like to teach to Rex:

1. Locomotion - Run/Walk
2. Stand up
3. Reach a specific point
4. Autonomous navigation - Map environment
5. Grab an object

Locomotion: Run

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 Sim-to-Real studies. 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 rex_reactive_env.py) 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 /policies/galloping/-++-rex_reactive.

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 policies/galloping/videos/rex-no-bounds.mp4

Galloping gait - bounded feedback

To improve the gait, in this second simulation, I've worked on the leg model.

I set bounds for both Leg and Foot angles, keeping the Shoulder in the initial position. I've ran the simulation (7M steps), the output control policy is in /policies/galloping/bounded-rex_reactive.

The emerged gait looks more clear. The policy video is policies/galloping/videos/rex-galloping.mp4

Credits

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.