Inverse reinforcement learning algorithms

Hi, I am working with rllib for some time now and really enjoying all the implemented algorithms, so thanks to all who worked on them. Right now I am interested in inverse reinforcement learning. Since I did not find any algorithms for that, I am wondering if I am just not looking at the correct pages or whether there are no implementations of inverse reinforcement learning yet? If the latter is the case, is there anything planned in this direction?

Hey @TSK93 . Thanks for asking this question!

We have indeed no inverse RL algos so far implemented in RLlib and are not planning to do so in the near future (not b/c of lack of interest but more due to lack of resources). If you would like to step in and provide an implementation, that would be super great! Happy to help with this in case there are questions about how to go about it. Also, of course, please consider making this a PR so everyone could benefit from your efforts.

Hi @sven1977,
thank you for the fast reply. I talked to a collegue of mine and we agreed to have at least a look on the topic and if we happen to get a nice implementation we would be glad to share it within the ray project then.

1 Like