OffWorld Gym is a collection of real-world environments for reinforcement learning in robotics with free public remote access. Close integration into the existing ecosystem allows you to start using OffWorld Gym without any prior experience in robotics. It removes the burden of managing a physical robotics system and abstracts it under a familiar API. Our first task is set up on analog lunar surface where a robot has to visually locate a physical object and approach it. As OffWorld Gym expands, additional environments will include a wide range tasks that reflect real-world challenges in industrial and space robotics. With your help we hope to accelerate development of the technology and the knowledge base that will enable humans to have a robust and capable robotic workforce in remote extreme environments, on Earth and in Space.

OffWorld Inc.

OffWorld Gym is operated and maintained by OffWorld Inc. OffWorld is developing a new generation of industrial robots to do the heavy lifting on Earth, Moon, asteroids and Mars. We see a new generation of industrial robots as the key enabler of human expansion beyond our home planet. Starting from scratch on the Moon or Mars is hard as neither of these places naturally sustains life as we know it. To survive, human settlers will need to endure deep space travel, battle harsh radiation, operate in lower gravity, and live in artificial atmosphere. Getting help from Earth could take weeks if you are on the Moon and years if you find yourself on Mars. We believe the best way to reduce the extraordinary risks involved in establishing permanent and sustainable presence on other planetary surfaces is to have a local robotic workforce to do the heavy lifting: build landing pads, excavate underground habitats, extract water ice and materials, make drinkable water, breathable air and rocket propellant, manufacture basic structures and solar cells, produce electricity, and eventually replicate themselves. Over the next decades, we will need thousands of robots that could mine, manufacture and build on the Moon, asteroids and Mars. All those robots need to be autonomous and adaptable to ever-changing conditions, we see reinforcement learning as one of major candidate technologies that will allows us to reach the required level of autonomy. The range of tasks we need to be able to perform robustly is vast. With the release of OffWorld Gym we look forward to engaging the community and accelerating learning for real-world industrial and space robotics.


Details on OffWorld Gym architecture, concept and experiments are available in our paper
"OffWorld Gym: open-access physical robotics environment for real-world reinforcement learning benchmark and research"

Please use the following BibTeX entry if citing OffWorld Gym in a publication

  Author = {Ashish Kumar and Toby Buckley and Qiaozhi Wang and Alicia Kavelaars and Ilya Kuzovkin},
  Title = {OffWorld Gym: open-access physical robotics environment for real-world reinforcement learning benchmark and research},
  Year = {2019},
  Eprint = {arXiv:1910.08639},


For all inquiries please reach out to gym.info@offworld.ai.
For the terms of use of OffWorld Gym please refer to Terms of Service.