Shreyas Gite
@shreyasgite
AI products at http://tuul.ai. Prev: founded self-driving at Kopernikus (acq. @Ford), jet engines @RollsRoyce.
Gemini + π0 = actually useful robots! (Similar to what @physical_int did with "Hi Robot") I can now verbally tell the robot that I'm building a red Lego wall or wooden tower, and it will infer the next steps by itself and pass me the necessary pieces, tools, or materials, ha!…
Someone should integrate eFlesh into the PincOpen gripper from @pollenrobotics and sell it as an accessory for @LeRobotHF so100/so101. It would sell like hot cakes; I would buy a couple of them.

My thoughts: For training the world model, you can use whatever data you collected during iterations and trials. For training the BC policy, you can only use high quality data. World model is like a trash recycling bin so you don't waste the data if it's not good enough.
And World Models?! JEPA v2, 1X Redwood, Nvidia... My shiny object syndrome kicked in hard. Tried poking open video models for physics-based grounding in robotics. Yeah... NOPE. World Models are seductive research territory but fucking expensive with zero clear robotics…
And World Models?! JEPA v2, 1X Redwood, Nvidia... My shiny object syndrome kicked in hard. Tried poking open video models for physics-based grounding in robotics. Yeah... NOPE. World Models are seductive research territory but fucking expensive with zero clear robotics…
My limited experience in robotic manipulation very well aligns with Sergey's blog. Things I did (that you shouldn't) if your goal is fast robot deployment. Simple goal: lego sorting robot. Had a working Gemini-Pi0 system, 1000 episodes collected March-May 2025. Should've doubled…
My limited experience in robotic manipulation very well aligns with Sergey's blog. Things I did (that you shouldn't) if your goal is fast robot deployment. Simple goal: lego sorting robot. Had a working Gemini-Pi0 system, 1000 episodes collected March-May 2025. Should've doubled…
I wrote a fun little article about all the ways to dodge the need for real-world robot data. I think it has a cute title. sergeylevine.substack.com/p/sporks-of-agi
🚀🚀🚀 Ever wondered what it takes for robots to handle real-world household tasks? long-horizon execution, deformable object dexterity, and unseen object generalization — meet GR-3, ByteDance Seed’s new Vision-Language-Action (VLA) model! GR-3 is a generalizable…
“We can only “fool” our models and prevent them from realizing that they are in The Matrix by lobotomizing them.” Not sure if we are talking about robots anymore?!😅
I wrote a fun little article about all the ways to dodge the need for real-world robot data. I think it has a cute title. sergeylevine.substack.com/p/sporks-of-agi
This video from Dyna, gives a good idea for distance between the bimanual robot arm. Also one more dof around wrist or elbow would be handy!
try moving the shoulders farther apart so the two arms are more parallel during the hand-off the should avoid the awkward angle that requires wrist bending to avoid
What is the German word for debugging for an hour trying to figure out why you cannot detect the motors, only to realize the power brick is not connected to the outlet? ...asking for a friend 👀
I underestimated how hard it is to have two robot arms interact. For ppl with more experience in the domain, what is the ideal way to set it up: - Move the camera on red arm to other side to avoid collision? - Should the arms rather be facing each other rather than side by side?…
So far I avoided wrist cameras on robots. I experimented a lot with image inpainting for data augmentation, and getting consistent image generation across different perspectives was already a hassle - didn't need the extra complexity of moving cameras. Recently learned that…

If productivity of your European colleagues with kids suddenly falls, that means the six week of school summer holidays have kicked in. See you all in September 😅
I still find it mysterious whether and how intelligence and capabilities transfer between domains and skills - from meta learning during early days to more recent question like whether solving maths helps writing a good essay. Sometime I feel a bit pessimistic given not enough…
Trying to hack together a Robot that can do things🤖 Bit limited by my CAD skills, anyone in Berlin/Potsdam that can help me out? Or is there like cursor for cad?

Robotics startup idea: Evaluation as a Service. In the LBM (Large Behavior Models) paper from TRI, the team did 47,000 evaluation rollouts in simulations. As Russ Tedrake emphasized: real-world evaluations are prohibitively expensive, and you can't trust loss functions alone.…

What are robot world models? If you follow the robotics space, you'll have almost certainly heard the term "world model." In the end, these mean using generative ai to build data-driven simulators or planners in order to build general-purpose robots. In this post I write about…