Pinned Tweet
Nathan Fielder claims that You Can See Everything.
Ruby, our RBY1 humanoid, thinks that’s a pretty strong observability assumption, so we taught her to Know Where to Look instead.
Check out our work: arxiv.org/abs/2611.00499
[8/n] SSF was a really fun project to work on with my amazing collaborators at the @MIT_LISLab at @MIT_CSAIL with @lucyjcai (equal contribution), @skyflylu, @MandersonTravis, Nicholas Roy, Tomás Lozano-Pérez, and Leslie Kaelbling. We also put a lot of thought into the project website - belief visualization, side-by-side runs, efficiency analysis, and failure modes - so please take a look! And come find us at the LEAP and UncertainWM workshops at #CoRL2026 #corl (@corl_conf).
website - 🌐 skymanaditya1.github.io/objs…
arxiv - 📄arxiv.org/abs/2611.00499
video - 📹youtube.com/watch?v=O7iGAn83…
kelly retweeted
Show a robot a cart it's never seen. It doesn't need to know it's a "cart," just that it's pushable, rolls & holds stuff.
Happy to announce my first-author paper on functional latent spaces got into #CoRL2026: "What Objects Enable, Not What They Are"
arxiv.org/abs/2606.05533
🧵
kelly retweeted
Imagine asking your robot to find your car keys🔑. They could be on your desk, the key holder, or in a drawer. The robot can't see most of the room - it has to reason about likely hiding spots, go look, and open things. Can robots do this today? Our new work, SSF, does exactly that! 🤖🧵
Ruby gave me back pain #CoRL2026
[5/n] We deploy this system on a bimanual wheeled humanoid, the RB-Y1🦾. To enable the robot to perform these tasks, we build a skill repository consisting of A*-based navigation, whole-body IK to place the head camera for a closer look, and whole-body pick-and-place skills for tables, shelves, drawers, and cabinets by adapting TipTop to the robot's torso and 7-DoF arms. For interacting with closed containers, we train cabinet and drawer-opening policies via behavior cloning with Diffusion Policy on 150 demonstrations.
kelly retweeted
Introducing MalwareBench: the first end-to-end, cross-platform benchmark on malware reverse engineering.
Most RE benchmarks are saturated, but not MalwareBench. GPT-6.1 Sol leads with 4.42%. A typical run takes several hours and hundreds of millions of tokens.
kelly retweeted
We are releasing the first liquid way to trade Anthropic.
For far too long, frontier assets have only been accessible to a select few. We are here to bring them to the masses.
Introducing EntropyIO. We’ve raised $14M led by Ribbit Capital and a $40M HYPE stake to build both improved and entirely new markets, the first of which are live on Hyperliquid today.
kelly retweeted
today we're launching automatic captioning at Instance.
send us your robot data, and every episode is segmented into captioned subtasks, graded success/fail plus 1-5 on quality and speed.
if this could be useful for you, reach out and we'll caption an episode for free!
kelly retweeted
Tech has decided that regulation is the enemy of progress, but innovation requires capital
Regulation builds the trust necessary to inject billions into markets
To enable free innovation, Andera raised a $37M series A led by Lightspeed to scale financial oversight with agents
kelly retweeted
Hosting a fireside chat with @AshwinSreenivas on building Decagon from 0->1. We'll talk about the early days, what worked, what didn't, and lessons from scaling the company. DM me for an invite!
