RL and world models @ Tesla Optimus | Previously: PhD @ UC Berkeley
Joined May 2021
- Tweets63
- Following562
- Followers122
- Likes515
Utkarsh Singhal (going to CoRL '26) retweeted
Pi models are now running in production settings, in collab with @Ultraroboticsco and @weaverobotics.
We see:
- much higher autonomy with pi-0.6 over using pi-0.5
- fewer mistakes & higher throughput from incorporating data in pre-training
Blog post: pi.website/blog/partner
Utkarsh Singhal (going to CoRL '26) retweeted
It was absolutely disgraceful seeing the venue people just rudely cut the mic. Even when @breadli428 repeatedly beg for just 3 more minutes and he will be done, the person simply repeated robotically: you should have been done. I was clearing rooms one by one. It’s the event organizer’s fault.
Absolutely no respect at all.
The toughest moment in a PhD:
>spend a year building smth you’re proud of
>travel across world with advisors’ support to share it
>when your moment finally comes
>your mic gets cut because the previous schedule ran late
Heartbroken, but thanks to whom stayed for me @NeurIPSConf
Happening now 😀 ICML poster session, E. Hall A-B (E-2203)
Q: How do we scale robustness/invariance to foundation models like CLIP?
A: Test-time search! 🔍
Our new work FoCal finds canonical views to boost robustness to complex transforms (e.g. viewpoint): sutkarsh.github.io/projects/…
📍 ICML Poster: Tue 11–1:30, E. Hall A-B (E-2203)
🧵 1/5
Check out our work @ ICML tomorrow!
I can't join in person 🥲, but I'm genuinely excited to share this work.
FoCal hits the core issues I've encountered working on invariance over the last few years: complex transforms, scaling to foundation models & data-driven invariance 😀
Q: How do we scale robustness/invariance to foundation models like CLIP?
A: Test-time search! 🔍
Our new work FoCal finds canonical views to boost robustness to complex transforms (e.g. viewpoint): sutkarsh.github.io/projects/…
📍 ICML Poster: Tue 11–1:30, E. Hall A-B (E-2203)
🧵 1/5
Utkarsh Singhal (going to CoRL '26) retweeted
Our gloriously picked cherries.
Their wicked failure cases.
Hi all, I'll be at NeurIPS from Tuesday to Sunday.
I would love to chat with people about invariance, test-time optimization, and vision.
Please DM me if you'd like to talk (or catch up)
Utkarsh Singhal (going to CoRL '26) retweeted
Possibly my favorite reaction to our work.
@utksinghal @_k_a_c_h_
Huh. Weird.
arxiv.org/abs/2312.04709
Utkarsh Singhal (going to CoRL '26) retweeted
🧵: How do you design a network that can optimize (edit, transform, ...) the weights of another neural network?
Our latest answer to that question: *Universal* Neural Functionals (UNFs) that can process the weights of *any* deep architecture.
Utkarsh Singhal (going to CoRL '26) retweeted
Sora is our first video generation model - it can create HD videos up to 1 min long. AGI will be able to simulate the physical world, and Sora is a key step in that direction. thrilled to have worked on this with @billpeeb at @openai for the past year openai.com/sora
Utkarsh Singhal (going to CoRL '26) retweeted
Can you guess Δw before you even know the loss?
Our new work, "How to guess a gradient"(arxiv.org/abs/2312.04709) shows you can make much better guesses than uniform random without the loss function.
Work co-led by @utksinghal, @_k_a_c_h_, and myself. (1/n)
ALT https://arxiv.org/abs/2312.04709
Excited to announce our ICCV’23 paper: “Learning to Transform for Generalizable Instance-wise Invariance”!
sutkarsh.github.io/projects/…
Why is it invariant? In short, even when the inputs have very different poses, the predicted distributions (in the absolute pose space) are much closer. So, the average predictions over them will be closer, too. We formalize this in our paper and show many more results.
Curious to learn more? Check out our poster 177 at 2:30-4:30 PM Wednesday (today)!
You can also check out our website: sutkarsh.github.io/projects/…