The Farama Foundation is a 501c3 nonprofit organization dedicated to advancing the field of reinforcement
learning through promoting better standardization and open source tooling for both researchers and industry.
Currently we do this through maintaining three widely used core libraries that offer standard APIs between reinforcement learning different kinds of “environments” (e.g. robotics simulations, video games, etc.), as well as many of the most widely used reinforcement learning environments for researchers. You can view the full list of our libraries here; they have 10s of millions of installations and are depended on by 10s of thousands of projects and represent many of the most installed reinforcement learning (or just machine learning in general) libraries in the world.
Our short term goals are to make various long planned large quality of life, productivity performance improvements to our core libraries and to promote their adoption throughout the open source reinforcement learning ecosystem, to provide maintenance and bug fixes to the environments we have that are mature, and to bring in more key open source reinforcement learning projects and bring them up to our standards for maturity. When we conducted an informal poll of 50 professional RL researchers in 2022, the average researcher reported that our planned quality of life and productivity improvements to these libraries will save them an average of 1 week per paper. Many of the environment upgrades that we're working on will hopefully result more an an a 10x speedup (though some will be 1000x or more).
Our long term goals involve expanding into offering better standardization tools for engineers and academics working on applied reinforcement learning tasks (though things like providing a Gymnasium-like standard C API), as well as to provide key open source tools for reinforcement learning beyond just the environment layer of the stack.
The team supporting this effort includes roughly 25 core members, and roughly 80 non-core contributors, spanning approximately 12 time zones and every background imaginable. Many of our core contributors came from the team behind OpenAI Gym, which we were the maintainers of for 2 years before it was replaced by our Gymnasium library.
If you'd like to read in more detail about our origins and motivations, please check out our announcement blog post. If you'd like to get in touch with us about funding or anything else, you can reach out at firstname.lastname@example.org, but our community is primarily organized through a large discord server. If you'd like to donate, you can do so here; the funds go to paying our engineers to work full time on these libraries.