Introduction
SofaEnv
The goal of this project is providing a Gym interface to SOFA in order to use SOFA as a physics-based environment for reinforcement learning.
The Simulation Open Framework Architecture (SOFA) is “an efficient framework dedicated to research, prototyping and development of physics-based simulations”. In particular its ability to perform fast and realistic soft body simulations make this framework appealing for reinforcement learning.
Gym environments provide a simple interface between reinforcement learning algorithms and learning environments. The basic components of a gym environment are their state and action spaces as well as a set of interface functions.
SofaEnv provides modular and simple to use functions to
define how actions, passed to the Gym environment affect what happens in the SOFA simulation, and
observe the state of the simulation.
Contents
If you are new to the project, please have a look at the Getting Started section.
The project consists of:
The SofaEnv Base Class that implements the Gym interface for SOFA simulations
Sofa Templates that combine SOFA components to model high-level objects such as deformable objects and pivotized instruments.
A set of predefined Scenes and their learning environments.
A set of Utility Functions to generate point cloud observations, perform motion planning, and more.