I'm a featured interview in our latest behind-the-scenes release! We break down the ML and perception that drives the whole-body manipulation behaviors from last year. It starts with a neat demo of Atlas's range-of-motion and our vision foundation models. piped.video/oe1dke3Cf7I?si=o0wM…

So snappy, and such an underrated bit of co-design!

In-context learning has always been a north star for robots, and seeing sparks of one-shotting is genuinely exciting!

Most world models for robot manipulation learn physics from pixels. But pixels don’t see it all. Can we ground these models in the "feeling" of contact to disambiguate visually identical states? Visuo-Tactile World Models (VT-WM): robot imagination in a shared space👇

Check out UMI-FT, the newest member of the UMI family! 🐣 We integrated a coin-sized🪙, low-cost, 6-axis force-torque sensor behind each UMI Fin Ray finger. The sensor is physically compact, mechanically robust, and extremely easy-to-calibrate against ATI -- directly measures physically meaningful force/torque outputs (N/Nm). UMI-FT enables in-the-wild robot learning with force-aware manipulation data, without needing in-the-wild robots or relying on traditional, bulky, thousands-dollar F/T sensors! A cross-paradigm bonus: because the coinFT sensor outputs physically meaningful force/torque values (N/Nm), you can directly run standard admittance control to get F/T-aware compliant robot fingers -- no learning required, immediately useful! Checkout @Hojung_Choi_' post for deets!

It's been amazing to work with @HaozhiQ and I've learnt a lot from him, excited for his new lab!

Same for ego data, UMI data, etc. An open secret is “we used ego4d” actually means filtering out the 1% of videos that are vaguely useful for learning. Occlusions, suboptimality, sensor noise, and so many pitfalls! Modeling + collect co-design is only gold standard currently.

I'm super excited to announce mjlab today! mjlab = Isaac Lab's APIs + best-in-class MuJoCo physics + massively parallel GPU acceleration Built directly on MuJoCo Warp with the abstractions you love.

I'll be giving a talk @CoRL2025 about developing large behavior models on Atlas. Come by the dexterous manipulation workshop at 11:30 to see the talk, dex-manipulation.github.io/c…

Can we scale up mobile manipulation with egocentric human data? Meet EMMA: Egocentric Mobile MAnipulation EMMA learns from human mobile manipulation + static robot data — no mobile teleop needed! EMMA generalizes to new scenes and scales strongly with added human data. 1/9

Lucas and co. wrote a great blogpost on the careful science and engineering behind language-conditioned policies for whole-body manipulation! There's a lot more work on the horizon; our team is hiring researchers to scale egocentric human data and VLMs for robotics. Reach out!

Viser is such an amazing tool for the community - congrats to @brenthyi and co!

TRI's latest Large Behavior Model (LBM) paper landed on arxiv last night! Check out our project website: toyotaresearchinstitute.gith… One of our main goals for this paper was to put out a very careful and thorough study on the topic to help people understand the state of the technology, and to share a lot of details for how we're achieving it. piped.video/watch?v=BEXFnru5…

Current robot policies often face a tradeoff: they're either precise (but brittle) or generalizable (but imprecise). We present ViTaL, a framework that lets robots generalize precise, contact-rich manipulation skills across unseen environments with millimeter-level precision. 🧵

Today we're excited to share a glimpse of what we're building at Generalist. As a first step towards our mission of making general-purpose robots a reality, we're pushing the frontiers of what end-to-end AI models can achieve in the real world. Here's a preview of our early results in autonomous general-purpose dexterous capabilities – fast, reactive, smooth, precise, bi-manual coordinated sensorimotor control.

How to learn dexterous manipulation for any robot hand from a single human demonstration? Check out DexMachina, our new RL algorithm that learns long-horizon, bimanual dexterous policies for a variety of dexterous hands, articulated objects, and complex motions.