Reinforcement Learning Framework

OpenAI’s Gym interface is a well-known interface in the Reinforcement Learning community, which allows to test many sequential learning models on problems ranging from robotics to video games. The general idea of this interface is to be able to interact with an environment, generally the simulation of an agent and its environment, from basic commands like:

• step(action) which allows to apply an action in the environment and which returns the information of the environment
• or render() which allows to display a visual rendering of the environment

The goal of SofaGym is to propose an open source software allowing to create a Gym environment, over any SOFA scene. This interface makes it possible to combine the simplicity of Gym with the computing power of SOFA, in order to test Reinforcement Learning or planning algorithms in the context of soft robotics and medical simulation. SofaGym also contains example scenes to show how everyone can create their own environment based on their particular problem. A baseline is also provided to compare different algorithms on simple soft-robotic examples.

RL for autonomous transthoracic catheter control

RL for soft robot control

Related publications

• SofaGym: An open platform for Reinforcement Learning based on Soft Robot simulations

Back to the plugin list