I'm at #RSS2024 in Delft this week - I'll be presenting my work on risk-sensitive RL (x.com/KyleStachowicz…) Wednesday and some new work on online RL+robot foundation models @ the lifelong learning workshop Friday! If you're at RSS and want to chat about RL, let me know!

Yet another year, yet another IEEE ICRA PaperCept server crash. Happy ICRA (not a) deadline to those who celebrate :)

Check out this work w/Carlo Sferrazza Oier Mees Josh Jones Sergey Levine Pieter Abbeel: fine-tuning VLA policies with multimodal inputs (vision, touch, audio) to generate sensory descriptions + robot actions! I’ve been pretty excited about VLA models lately, more to come soon :)

Excited to release FAST, our new robot action tokenizer! 🤖 Some highlights: - Simple autoregressive VLAs match diffusion VLA performance - Trains up to 5x faster - Works on all robot datasets we tested - First VLAs that work out-of-the-box in new environments! 🧵/

Please check out Karl Pertsch’s thread for more details! This was a collaboration led by Karl and Kyle Stachowicz, with help from many folks at Physical Intelligence x.com/KarlPertsch/st…

was pretty wild seeing the droid model work zero shot irl - first signs of actual robustness

Excited to announce that we'll be presenting FuSe at ICRA 2025!

R1's RL findings are great news for reasoning but grim for robotics. All the major takeaways (ground-truth reward, great base models, grouped rollouts from same initial state, sample-inefficient on-policy algos) are really hard to translate to the physical world.

The π0-FAST model is also included in this release (both base and a model finetuned for DROID), check it out if you've got access to a DROID setup!

Excited to play with Pi0, it is so cool! We just config DROID, download ckpt & inference code, and it works without any tuning, what a impressive moment

Check out what we've been working on at π! x.com/physical_int/s… - zero-shot generalization scales with diverse data (!!!) - FAST pre-training+diffusion fine-tuning gives the best of both worlds And lots more - check out our blog post and paper!

And we're hiring! Apply on our website or DM me with questions

Exciting workshop from Will Chen and friends, take a look!

We've added pi-05 to the openpi repo: pi05-base, pi05-droid, pi05-libero. Also added PyTorch training code!🔥 Instructions and code here: github.com/Physical-Intel… This is an updated version of the model we showed cleaning kitchens and bedrooms in April: pi.website/blog/pi05