Soft-Robot Control Environment (gym-softrobot)

About

Soft robots replace conventional rigid links and localized joints with continuously deformable bodies. Their compliance can make them safe, adaptable, and capable of rich motion, but it also makes control and actuator design difficult: deformation is nonlinear, actuation is often distributed throughout the body, and an actuator changes the mechanics of the structure that it is trying to control. Useful control studies therefore need both realistic continuum dynamics and freedom to ask how the robot should be actuated.

gym-softrobot provides Gymnasium environments for this problem, using PyElastica for high-fidelity Cosserat-rod simulation. Rather than committing every task to a single hardware design, the environments cover several levels of actuator abstraction:

• direct push-pull forces for simple control experiments;
• tendon-driven actuation for continuum arms;
• distributed, muscle-like activation for bio-inspired robots; and
• idealized force and torque controls for studying hypothetical actuator placement and authority.

This range makes it possible to separate control questions from actuator-design questions, then progressively introduce physical constraints. The environments can be used to investigate policy transfer between actuation models, curriculum learning from simpler to higher-fidelity tasks, sim-to-real-oriented robustness, co-design, and reinforcement-learning methods for distributed control.

Installation

gym-softrobot requires Python 3.12 or newer.

pip install gym-softrobot

Quick start

import gymnasium as gym
import gym_softrobot  # Registers the environments.

env = gym.make("SoftPendulum-v0")
observation, info = env.reset(seed=42)

terminated = truncated = False
while not (terminated or truncated):
    action = env.action_space.sample()
    observation, reward, terminated, truncated, info = env.step(action)

env.close()

For a more involved example, the CyberOctopus environments expose distributed feedback control over multiple soft arms:

import gymnasium as gym
import gym_softrobot

env = gym.make("OctoMuscleCrawl-v0")
observation, info = env.reset(seed=42)
observation, reward, terminated, truncated, info = env.step(
    env.action_space.sample()
)
env.close()

See the documentation for the environment catalog, API conventions, rendering, CyberOctopus variants, and RL examples. The registered IDs can also be listed with:

python -m gym_softrobot.debug.registry

Development

uv sync --all-groups
uv run pre-commit install

Citation

@misc{gym_softrobot,
  author = {Seung Hyun Kim and Chia-Hsien Shih and Mattia Gazzola},
  title = {Soft Robotics Environment for Gymnasium},
  year = {2026},
  publisher = {GitHub},
  howpublished = {\url{https://github.com/skim0119/gym-softrobot}}
}

@article{Naughton2021,
  author = {Naughton, Noel and Sun, Jiarui and Tekinalp, Arman and
            Parthasarathy, Tejaswin and Chowdhary, Girish and Gazzola, Mattia},
  title = {Elastica: A Compliant Mechanics Environment for Soft Robotic Control},
  journal = {IEEE Robotics and Automation Letters},
  year = {2021},
  volume = {6},
  number = {2},
  pages = {3389--3396},
  doi = {10.1109/LRA.2021.3063698}
}

Author