Excited to see Jensen share our recent progress on the N1.5 foundation model and GR00T Dreams from the NVIDIA GEAR team! Several team members and I will be at #ICRA2025. Come chat with us about building generalist robot autonomy and ways to work together on this grand mission!

Just finished this talk on radiance fields and generative media from Jon Barron One of the more honest assessments I’ve seen about where 3D actually matters and where it doesn’t. Turns out you don’t need explicit 3D to get photorealism (see: Sora), or to control a robot

Products with extensive/rich UIs lots of sliders, switches, menus, with no scripting support, and built on opaque, custom, binary formats are ngmi in the era of heavy human+AI collaboration. If an LLM can't read the underlying representations and manipulate them and all of the

My sleep scores during recent travel were in the 90s. Now back in SF I am consistently back down to 70s, 80s. I am increasingly convinced that this is due to traffic noise from a nearby road/intersection where I live - every ~10min, a car, truck, bus, or motorcycle with a very

A lot of interesting things in that Helix demo: First is obv the scaling of the dataset. It's not just raw scaling from 10 to 60 hrs; there's more to it. - Monocular to stereo vision: If you wanted to do this with LeRobot, training on RGB-D or point clouds would require a

I’ll be at CVPR in Nashville this week, for those in my network who may be interested in connecting. Feel free to send me a DM to coordinate!

In my spare time a few weeks ago, I worked on something fun! I trained the GPT model series from scratch (like everyone else xD) No pre-trained weights and no shortcuts: just pytorch, cuda, and an 8 gpu node (thanks Lambda). It's always fun to implement the foundations

Good substack post by Sergey Levine (linked below) Training large-scale robotic agents is hard and gets harder the further we drift from real-world data. Simulations, human videos, hand-held grippers all sporks 🥄🍴, according to Sergey. I would tend to agree as they create narrow

Cool to see folks with similar thoughts. I’m still of the opinion that simulation and other surrogate methods have a place in VLA etc pre training, but maybe less applicability when going from 0 -> 1 on a task

Just watched an inspiring Chelsea Finn talk at YC on scaling physical intelligence. Coolest takeaway for me was on her points on building high-level policies for better natural language instruction following. I see several interesting research directions (if not already

Think & act is the new trend of robotics generalists model!

Come hear from Alexander Soare from Cobot present his work 'Smooth-as-Butter Robot Policies," building on Real-Time Action Chunking from Physical Intelligence! alexander-soare.github.io/robotics/2025/…

Such exciting times in the field of Physical AI. This last week at the Humanoid Summit, leaders from across the world of Physical AI convened to discuss the state of the industry and what is coming next. My main takeaway leaving the conference -> the industry is not there yet by

Why do generalist robotic models fail when a cup is moved just two inches to the left? It’s not a lack of motor skill, it’s an alignment problem. Today, we introduce VLS: Vision-Language Steering of Pretrained Robot Policies, a training-free framework that guides robot behavior

What if robots could think longer on harder problems without saying a single word?🤔 We introduce RD-VLA (Recurrent-Depth VLA): a latent, iterative reasoning architecture for robot control. ❌No Chain-of-Thought tokens. ❌No extra memory overhead. ✅Just reasoning—directly in