Murtaza Dalal
@mihdalal
AI @ Tesla Optimus | Robotics/AI PhD @CMU_Robotics, EECS @UCBerkeley | Opinions entirely and unequivocally my own
I want to make clear how crazy impressive this result is. We can now do bi-manual, dexterous manipulation across a wide range of tasks with barely any data on these skills coming from teleoperation. As we know, teleop does not scale! But turns out human video does! This means…
I’m not just dancing all day, ok
This is a significant leap forward for grasping! Scaling in sim is definitely the right approach this problem.
I’m thrilled to announce that we just released GraspGen, a multi-year project we have been cooking at @NVIDIARobotics 🚀 GraspGen: A Diffusion-Based Framework for 6-DOF Grasping Grasping is a foundational challenge in robotics 🤖 — whether for industrial picking or…
Gotta get the power of the sun in my hand iykyk ☀️
The math is obvious and unequivocal: ~100% of energy over time comes from the Sun. Everything else is a rounding error.
Cool work by Mihir and co! Will be interesting to see if these findings hold at frontier scale
🚨 The era of infinite internet data is ending, So we ask: 👉 What’s the right generative modelling objective when data—not compute—is the bottleneck? TL;DR: ▶️Compute-constrained? Train Autoregressive models ▶️Data-constrained? Train Diffusion models Get ready for 🤿 1/n
Robot scifi future is upon us Baby steps 🍿
FULL interaction with Optimus at Tesla Diner in Hollywood! Had a chance to talk to one of the engineers on the team, he said “they are progressing very quickly every week” Imagine coming to this place and this is what greets you for your orders 🤯🤯🤯
Bio + AI is the future and evaluation of models aka benchmark is the key to accelerating progress. Check out this great work from @Taykhoom_Dalal that builds up a comprehensive suite for evaluating mRNA foundation models! Definitely worth a read for those interested in getting…
I am excited to introduce mRNABench, a comprehensive benchmarking suite that we used to evaluate the representational capabilities of 18 families of nucleotide foundation models on mature mRNA specific tasks. Paper: doi.org/10.1101/2025.0… Code: github.com/morrislab/mRNA… A 🧵
🦿🦾
That’s Optimus 2.5. Optimus 3 will have agility roughly equal to an agile human.
Welcome @bingjietang07! Very excited to have you here with us to build the humanoid robotics future!
Excited to share I’m joining @Tesla_Optimus! Huge thanks to my advisor, @gauravsukhatme, labmates @uscresl, collaborators @NVIDIARobotics, and all the friends and family who supported me throughout my PhD.
This is what we get the PhD for, FINALLY I get to use the flex Let's hope if something goes wrong and they ask is a doctor on board, people don't look for me...
Full self delivering
World's first autonomous delivery of a car! This Tesla drove itself from Gigafactory Texas to its new owner's home ~30min away — crossing parking lots, highways & the city to reach its new owner
Incredibly simple method to induce novel task solving during inference time: warm start denoising with human priors!
Presenting DemoDiffusion: An extremely simple approach enabling a pre-trained 'generalist' diffusion policy to follow a human-demonstration for a novel task during inference One-shot human imitation *without* requiring any paired human-robot data or online RL 🙂 1/n
Taking “teleop” to the next level 🧠 🤖 Jk but like very cool result didnt think it was possible to do this
Watch the latest update from the Neuralink team.
As someone who has often just done this based intuition, this would be super useful in general ngl Would love to see a slightly more formal treatment
I wish there was a good venue to write/present about RL “tricks”. how PD gains affect action scale, how to tune reward functions, actor STD, etc. there’s good intuition for all of it, grounded both in learning theory and robot dynamics, but I don’t often see good explanations.
Hopefully one day we will have FSD deliver Optimus to customer doorsteps too :P Or a legion in a Tesla Semi, that would be cool
The first fully autonomous delivery of a Tesla Model Y from factory to a customer home across town, including highways, was just completed a day ahead of schedule!! Congratulations to the @Tesla_AI teams, both software & AI chip design!
🤖 🚀
Optimus V3 is going to be 🤌 Excellent review w @Tesla_Optimus team
Great work by AP to ship this officially today :) Can’t wait for it to come to the Bay
tesla.com/robotaxi
Active perception is a critical skill for robotic manipulation!
Your bimanual manipulators might need a Robot Neck 🤖🦒 Introducing Vision in Action: Learning Active Perception from Human Demonstrations ViA learns task-specific, active perceptual strategies—such as searching, tracking, and focusing—directly from human demos, enabling robust…
Yes!! Thrilled to present Neural MP at IROS🥳. We’ve also been pushing this line of work even further! Stay tuned, big things coming soon😜
Incredibly excited to share that Neural MP got accepted to IROS as an Oral presentation!! Huge congrats to the whole team (@Jiahui_Yang6709, @mendonca_rl , youssef khaky, @rsalakhu, @pathak2206), but especially @Jiahui_Yang6709 for making this happen after I graduated! This now…
Cool work on neural motion planning. much more challenging problem in a lot of ways than vision-language models - because you have to be able to come up with paths between any pair of configurations. good general purpose tool.
Incredibly excited to share that Neural MP got accepted to IROS as an Oral presentation!! Huge congrats to the whole team (@Jiahui_Yang6709, @mendonca_rl , youssef khaky, @rsalakhu, @pathak2206), but especially @Jiahui_Yang6709 for making this happen after I graduated! This now…
Incredibly excited to share that Neural MP got accepted to IROS as an Oral presentation!! Huge congrats to the whole team (@Jiahui_Yang6709, @mendonca_rl , youssef khaky, @rsalakhu, @pathak2206), but especially @Jiahui_Yang6709 for making this happen after I graduated! This now…
Can a single neural network policy generalize over poses, objects, obstacles, backgrounds, scene arrangements, in-hand objects, and start/goal states? Introducing Neural MP: A generalist policy for solving motion planning tasks in the real world 🤖 1/N