Everyone can now train high-dimensional muscle agents with RL, as we just released our code! Come join me at the #MyoChallenge workshop tomorrow at 10:00 in room 354 at #NeurIPS2023 to see my poster. Code is here!

Thanks to @tgeijten for his tremendous help with SCONE and Hyfydy, making the integration with deprl even better. I‘m proud to announce that we have created RL environments together that allow anyone to use Python with RL and Hyfydy.

Our adaptive energy cost term works seamlessly with #MyoSuite and with Hyfydy. All other cost terms are already included in sconegym and will follow to MyoSuite in a future update. Thanking @Vikashplus and @CaggianoVitt for their invaluable expertise with MuJoCo

If you enjoy seeing researchers open-source their software, please give us a star on GitHub And please read the docs!

Many works propose some RL control for muscles, but we managed to create natural looking and robust behaviors without learning from demonstrations. The adaptive cost terms we propose are enough to learn natural motion across 4 different models without changing parameters.

@tgeijten @HaeufleLab @GMartius @CaggianoVitt @Vikashplus I feel so sad for this guy every time I see his little armless run.

@tgeijten @HaeufleLab @GMartius @CaggianoVitt @Vikashplus Attack on Titan last season fell off hard

@tgeijten @HaeufleLab @GMartius @CaggianoVitt @Vikashplus Impressive work, but isn't this how we get the AI uprising?

We also train (torture) lower-dimensional models, producing robust gaits, without changing any hyperparameters! It all works with the same settings!

@tgeijten @HaeufleLab @GMartius @CaggianoVitt @Vikashplus It me

@tgeijten @HaeufleLab @GMartius @CaggianoVitt @Vikashplus FIFA needs to add this to their games