rex-gym 0.1.7

No project description provided

Rex-gym: OpenAI Gym environments for a quadruped robot

This repository contains different Gym Environments used to train Rex, the Rex URDF model, the learning agent and some scripts to start the training session and visualise the Control Polices.

Rex: an open-source domestic robot

The goal of this project is to train an open-source 3D printed quadruped robot exploring Reinforcement Learning and OpenAI Gym. The aim is to let the robot learns domestic and generic tasks in the simulations and then successfully transfer the knowledge (Control Policies) on the real robot without any other manual tuning.

This project is mostly inspired by the incredible works done by Boston Dynamics.

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 rex-gym package:

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 -m rex_gym.playground.rex_reactive_env_play

Check out the recorded video files 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 

Where YOUR_LOG_DIR_PATH is the output policy path.

Choose the Gym Environment setting the --config flag:

Task	Flag
Run	rex_reactive
Walk	rex_walking

PPO Agent configuration

You may want to edit the PPO agent's default configuration, especially the number of parallel agents launched during the simulation.

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

def default():
    """Default configuration for PPO."""
    # General
    ...
    num_agents = 20

Install rex_gym from source. This configuration will launch 20 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.

Simulation model

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 the list of basic tasks I'd like to teach to Rex:

1. Locomotion - Run/Walk
2. Stand up - Falling recovery
3. Reach a specific point in a map
4. Grab an object

Locomotion: Run

This task is about let Rex learns how to run on a straight line.

Gym Environment

There is a good number of papers on quadrupeds locomotion, most of them with sample code. Probably, the most complete collection of examples is the Minitaur folder in the Bullet3 repository. For this task, I've edited the Minitaur Reactive Environment explained in the paper Sim-to-Real: Learning Agile Locomotion For Quadruped Robots.

Galloping gait - from scratch

In this very first experiment, I let the system learn from scratch: giving 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).

The emerged galloping gait shows the chassis tilled up and some unusual positions/movements (especially starting from the initial pose) during the locomotion. The leg model needs improvements.

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.

The emerged gait now looks more clear.

Galloping gait - balanced feedback

Another test was made using a balanced feedback:

The Action Space dimension is equals to 4, assigning the same angle to both the front legs and a different one to the rear ones. The very same was done for the foot angles.

The simulation score is massively improved (about 10x) as the learning time while the emerged gait is very similar to the bounded feedback model. The Tensorflow score with this model, after ~500k attempts, is the same after ~4M attempts using any other models.

Locomotion: Walk

This task is about let Rex learns how to walk on a straight line.

Gym Environment

Starting from the Minitaur Alternating Leg example, I've used a sinusoidal signal as leg_model alternating the Rex legs during the locomotion. The feedback component has small bounds [-0.1,0.1] as in the original script, the sinusoidal function and the legs poses were edited.

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 in rendering the robot platform done by the SpotMicroAI community.