Skip to content

Latest commit

 

History

History
46 lines (29 loc) · 1.66 KB

README.md

File metadata and controls

46 lines (29 loc) · 1.66 KB
Raw

Status: I've just started working on the project again. Breaking changes might happen during the following months. I'm currently working on getting the framework ready for use as it was back in 2019, but with a completely refactored codebase.

Loco: A locomotion framework for RL research

Build Status

gif-ant-sample

gif-demo-sample

This is a locomotion framework that is intended to provide a new set of benchmarks for RL research in locomotion by allowing the user to create a wide range of diverse and complex environments. The core framework is engine-agnostic (it is not coupled to a specific physics engine), and allows to support a wide variety of physics engine. For now it support MuJoCo, and I'm currently working in support for Bullet as well (and PhysX in approx. July/August).

Setting up the project

Comming soon

Usage (Outdated)

Below is a minimal description of how to use the framework. I will be adding more documentation and moving around various components of the API. A simple example for the Ant-v2 environment (similar to Gym) consists of the following script:

import numpy as np

from loco.suite import basic

_env = basic.load( 'humanoid', 'walk' )

while True :
    _u = -1.0 + 2.0 * np.random.random( ( _env.actionDim(), ) )

    _env.step( _u )
    _env.render()

Some more examples can be found here.