OffWorld Gym: open-access physical lunar analog environment for reinforcement learning and robotics research
Abstract
Robotic workforce is likely to play a key role in the task of lunar and Martian settlement. The range of jobs robots can undertake includes construction activities, infrastructure deployment and repair, mining and in-situ resource utilization, digging habitats, in-situ material processing and more. Such robotic workforce will require high degree of autonomy to facilitate the full range of activities round the clock. In the last 5 years the field of Reinforcement Learning (a branch of machine learning and artificial intelligence) had demonstrated remarkable achievements in controlling virtual and simulated agents learning to operate autonomously and robustly in various environments. We think that this technology is ready to be taken out of simulators and applied to industrial robotic tasks in the real world, particularly for space settlement, where high degree of autonomy is highly desired. To facilitate research and progress in this direction we have created OffWorld Gym - a physical lunar analog environment with pre-configured industrial robotics tasks that will serve as a benchmark to track the progress in applying reinforcement learning to real physical systems. Success stories of applied machine learning can be traced back to the datasets and environments that were put forward as challenges for the community. The challenge that the community sets as a benchmark is usually the challenge that the community eventually solves. The ultimate challenge of reinforcement learning research is to train real agents to operate in the real environment, but until now there has not been a common real-world RL benchmark. Our collection of real-world environments for reinforcement learning in robotics has free public remote access, is closely integrated into existing ecosystem and allows the community to start using OffWorld Gym without any prior experience in robotics, taking away the burden of managing a physical robotics system. In the first iteration of the system we introduce a navigation task, where a robot has to reach a visual beacon on an uneven terrain using only the camera input and provide baseline results using state of the art reinforcement learning algorithms in both the real environment and the simulated replica. This platform will help evaluate applicability of reinforcement learning to planetary space robotics and guide the community efforts towards the tasks that are the most important for this field. Visit https://gym.offworld.ai to find out more.
- Publication:
-
43rd COSPAR Scientific Assembly. Held 28 January - 4 February
- Pub Date:
- January 2021
- Bibcode:
- 2021cosp...43E.164K