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.

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).

Comming soon

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.