[Question] Adding new tasks

Farama-Foundation / D4RL

A collection of reference environments for offline reinforcement learning

Apache License 2.0

1.32k stars 281 forks source link

[Question] Adding new tasks #169

Open albertcthomas opened 2 years ago

albertcthomas commented 2 years ago

Sorry if I missed this information but is it possible to add new tasks to this repo/benchmark?

pseudo-rnd-thoughts commented 2 years ago

Hi, the farama foundation have recently taken over the project and would be open to accepting new interesting environments. Could you contact me in the gym discord (https://discord.gg/nHg2JRN489) PseudoRnd and we can discuss more

kegl commented 1 year ago

@pseudo-rnd-thoughts pls add a new invite link, this has expired. I'm a colleague of @albertcthomas and we'd like to propose light D4RL tasks set where we have eg a 100 single episodes, each run with a different seed, so algorithms can be tried on extremely small data, but verified statistically properly across seeds. There are other variants between this and the extreme large data currently on D4RL. We'd like to discuss with other users to see if a consensus could be reached.

WillDudley commented 1 year ago

@kegl

https://discord.gg/bnJ6kubTg6

By all means feel free to generate more D4RL datasets. However, we are working on a superset of the D3RL dataset class known as Minari. This upgrade aims to address some reproducibility issues and API changes.

Please reach out to me via discord in the #d4rl channel.