trackmania-rl/tmrl

Still working on this project?

Closed this issue · 5 comments

glederrey commented

Hey,

I was wondering if you're still working on this project? I'd love to contribute to it if it's possible in any way. =)

(I'm a Data Scientist who worked on GANs during my PhD and I'd love to learn about reinforcement learning!)

yannbouteiller commented

Hello!

Sorry for the late answer, I'm on a trip an will be back early January. For sure we are still working on the project, version 0.4.0 is being developped on the yb/jax branch and should be released soon.

In particular it will introduce Internet security via the tlspyo library that we cut out of tmrl and released recently, stop forcing people to use PyTorch if they want to use Jax or something else, and hopefully "officially" launch our little self-racing competition in TrackMania.

If you like to help in any way you are very welcome!

(PS: closing as this is not an issue, please use the discussions or feel free to continue here)

yannbouteiller commented

If you want to learn deep RL and are a GAN expert maybe you might like to implement better CV-based approaches than the current vanilla CNN model and the current SAC/REDQ algorithm (which will be the point of the competition BTW)? Just an idea =)

yannbouteiller commented

Note that a cool thing about our little application here is that we are dealing with the real world: you will not only have to learn deep RL the MuJoCo way, but also to apply it to a real application with time constraints.

In particular, hugely computational models won't work at inference time.

glederrey commented

Hey Yann,

Thanks a lot for the answer. =) Do you have a discord where you dev? (I guess you're not the only one working on it ^^) If it's the case, I'd love to join it to discuss it in more details. =)

You can add me Tokay#7909 =)

yannbouteiller commented

We don't have a dedicated discord yet but people have asked before and that sounds like a good idea, maybe we should create one =)

  • 👍2