@MaxLobovsky Yep this is really nice to see. Big e2e models get all the attention, but work like this shows how far you can get with modern vision and a little controls elbow-grease. Ship it!
English
Jonathan Scholz
152 posts
This is the most promising direction in robotics today.
Problem: industrial robot arms are too expensive to install
Solution: low cost arms + vision + targeted machine learning technoques to make programming easier and system more robust to variation without lots of hardware engineering
Yunho Kim@awesomericky99
We present "hybrid system" that supplements conventional automation with "learning" for task & safety-level adaptiveness Deployed in factory for motor cable soldering (< 0.6 mm tolerance), resulting 108 motors, 99.4% SR with < 20 min data per task Paper: arxiv.org/abs/2604.22235
English
Eka means unity -- “one,” in Sanskrit and “first” in Finnish.
We’re building intelligence for the physical world in its native language: forces.
Until now, robotics faced a tradeoff — generality or speed. The real world requires both. Robotics also faced a data problem.
Our Vision–Force–Action (VFA) model — the first of its kind — breaks the generality-speed tradeoff and the data barrier.
It's a new foundation uniting performance, generality, and safety for putting capable robots in everyone's hands.
Today, I am excited to share our journey of pushing robots beyond human limits.
Today, dexterity becomes scalable.
Today, I welcome you to the Era of Eka.
Co-founded with @haarnoja, and so thrilled and grateful to be working with a dream team at @EkaRobotics.
Learn more: ekarobotics.com
English
*this* is what people should be excited about in robotics. Not skill demos - people using the platform for things the designers didn't anticipate.
Ben Pekarek@ben_pekarek
Today marks the end of my first full week @GeneralistAI Last Monday, I was given a challenge: use our GEN-1 model to teach a robot a task of my choosing, using the same no-code platform our customers use. I picked the ball-and-vase magic trick. It was one of my favorites as a kid, and it felt like the right mix of fun and surprisingly hard. A few days later, GEN-1 pulled it off. I left Friday having watched the robot nail it 14 times in a row. What’s wild is that even 4 months ago, if you told me you could go from idea to on-robot skill in a couple of days, I probably wouldn’t have believed you. Really excited to be building with an incredible team. Can’t wait to see what week two brings 🤖
English
This is great news for the (already great) London AI scene
Seb Johnson@SebJohnsonUK
Anthropic has announced that it is massively expanding its London presence. It’s just secured a new office for 800 people - a huge jump from its 200 current employees. OpenAI announced its first permanent office in London this week and now @AnthropicAI is doubling down. Meta, OpenAI, DeepMind, wayve and so many others have huge offices in London. It’s becoming the leading AI hub outside of the US. LETS GO
English
Really nice results, really nice analysis
steerability is *the* missing thing needed to get robotics outside the lab. Even more important than generalisation (though not unrelated)
Lucy Shi@lucy_x_shi
1/ We just released π0.7 — a steerable generalist robot model with emergent capabilities. I want to share a bit of the backstory, because π0.7 taught me something surprising about where robot learning is heading. A thread on bittersweet lessons 🧵
English
Jonathan Scholz retweetledi
The power of the Claw, in the palm of a robot hand. Agentic robotics is here! Today, we open-source CaP-X: vibe agents, alive in the physical world. They incarnate as robot arms and humanoids with a rich set of perception APIs, actuation APIs, and auto synthesize skill libraries as they go. CaP-X is a strict superset of our old stack, because policies like VLAs are “just” API calls as well. It solves many tasks zero-shot that a learned policy would struggle with.
And we are doing much more than vibing. CaP-X is our most systematic, scientific study on agentic robotics so far:
- We build a comprehensive agentic toolkit: perception (SAM3 segmentation, Molmo pointing, depth, point cloud), control (IK solvers, grasp planner, navigation), and visualization (EEF, mask overlays) that work across different robots.
- CaP-Gym: LLM’s first Physical Exam! 187 manipulation tasks across RoboSuite, LIBERO-PRO, and BEHAVIOR. Tabletop, bimanual, mobile manipulation. Sim and real. Can’t wait to see the gradients flow from CaP-Gym to the next wave of frontier LLM releases.
- CaP-Bench: we benchmark 12 frontier LLMs/VLMs (Gemini, GPT, Opus, Qwen, DeepSeek, Kimi, and more) across 8 evaluation tiers. We systematically vary API abstraction level, agentic harness, and visual grounding methods. Lots of insights in our paper.
- CaP-Agent0: a training-free agentic harness that matches or exceeds human expert code on 4 out of 7 tasks without task-specific tuning.
- CaP-RL: if you get a gym, you get RL ;). A 7B OSS model jumps from 20% to 72% success after only 50 training iterations. The synthesized programs transfer to real robots with minimal sim-to-real gap.
3 years ago, our team created Voyager, one of the earliest agentic AI that plays and learns in Minecraft continuously. Its key ideas — skill libraries, self-reflection loops, and in-context planning — have since influenced many modern agentic designs.
Today, the agent graduates from Minecraft and gets a real job. It’s April Fool’s, but this Claw is getting its hands dirty for real!
Link in thread:
English
Nice to finally see a worthy counterpoint to VLA in the language-conditioned robotics space!
I'm personally quite bullish on CLAW approaches, which lean on more sophisticated agentic cores than even the best VLAs.
This'll spark lots of ideas on the executive->control interface
Jim Fan@DrJimFan
The power of the Claw, in the palm of a robot hand. Agentic robotics is here! Today, we open-source CaP-X: vibe agents, alive in the physical world. They incarnate as robot arms and humanoids with a rich set of perception APIs, actuation APIs, and auto synthesize skill libraries as they go. CaP-X is a strict superset of our old stack, because policies like VLAs are “just” API calls as well. It solves many tasks zero-shot that a learned policy would struggle with. And we are doing much more than vibing. CaP-X is our most systematic, scientific study on agentic robotics so far: - We build a comprehensive agentic toolkit: perception (SAM3 segmentation, Molmo pointing, depth, point cloud), control (IK solvers, grasp planner, navigation), and visualization (EEF, mask overlays) that work across different robots. - CaP-Gym: LLM’s first Physical Exam! 187 manipulation tasks across RoboSuite, LIBERO-PRO, and BEHAVIOR. Tabletop, bimanual, mobile manipulation. Sim and real. Can’t wait to see the gradients flow from CaP-Gym to the next wave of frontier LLM releases. - CaP-Bench: we benchmark 12 frontier LLMs/VLMs (Gemini, GPT, Opus, Qwen, DeepSeek, Kimi, and more) across 8 evaluation tiers. We systematically vary API abstraction level, agentic harness, and visual grounding methods. Lots of insights in our paper. - CaP-Agent0: a training-free agentic harness that matches or exceeds human expert code on 4 out of 7 tasks without task-specific tuning. - CaP-RL: if you get a gym, you get RL ;). A 7B OSS model jumps from 20% to 72% success after only 50 training iterations. The synthesized programs transfer to real robots with minimal sim-to-real gap. 3 years ago, our team created Voyager, one of the earliest agentic AI that plays and learns in Minecraft continuously. Its key ideas — skill libraries, self-reflection loops, and in-context planning — have since influenced many modern agentic designs. Today, the agent graduates from Minecraft and gets a real job. It’s April Fool’s, but this Claw is getting its hands dirty for real! Link in thread:
English
Nice to see someone taking a theoretical position on intelligence that diverges from "RL is the definition and solution for everything"
I love me some RL, but human intelligence feels much more mimetic to me. It lives in the culture
Pedro A. Ortega@AdaptiveAgents
Agency is usually formalized as utility maximization. But must it be? LLMs suggest a different foundation: intelligence as acquiring behavioral schemas from interaction structure. My new paper: "Universal AI as Imitation" investigates the limit-case of LLM-style models.
English
MuJoCo 3.5 is out, including the official MJWarp release, a new sysID toolbox and lots of other goodies!
github.com/google-deepmin…
English
This is a narrative-breaking observation that should have people sitting up in their chairs
The moral isn't "give up on pretraining", its "find a way to train without losing too much rank"
(the language poeple got this right - step 1 shouldn't be "train on task data")
Tony Zhao@tonyzzhao
Maybe the bitter lesson for robotics that goes against all the current narratives is that a small amount of in-domain demos is really hard to beat, especially when it comes to e.g. industrial use cases, where variability is limited.
English
@tonyzzhao yep, you sacrifice too much steerability to get generality.
you had the right idea with ACT :)
English
Maybe the bitter lesson for robotics that goes against all the current narratives is that a small amount of in-domain demos is really hard to beat, especially when it comes to e.g. industrial use cases, where variability is limited.
Alper Canberk@alpercanbe
i was visiting a hackathon where 80+ participants were training pi0/0.5, gr00t, smolvla, ACT, DP, etc. on lerobot arms the best and most sample efficient policies were trained *from scratch* we still do not have an open source x-embodied GPT-2, but i'm hopeful for this year
English
i realised i really like kid shows 🤣
For intellectual stuff appreciate nuance, but for food and art and media i need to be hit over the head.
share.google/ZZBOqeHYheEv2d…
The themes in this show are super exaggerated, like the high fructose corn syrup of life lessons, and i love it
English
That’s the only way i know to make a team greater than the sum of its parts
thank you for coming to my TED talk
English
disagree? great! take your gloves off, so i can take mine off too, and we’ll get to the bone faster
but lets not try win the fight, lets try to win the peace. The atoms of debate should be ideas, not people.
English
I finally figured out why I think “disagree and commit” is bullshit:
It makes teams mode-seeking, not mean-seeking. it’s the last bastion of the fixed-minded.
my style is more like “disagree and converge”
English
in case i haven't said recently, i freaking love rerun.
this is gonna be super useful at R2, nice work @JafarUruc
Jafar Uruç@JafarUruc
Cleaned up the code and added simulation flag to the cli entrypoint 🤖🤖🤖
English
@ATOM1Humanoid @AgilexRobotics Nope, QDD not servos. Way more dynamic and interactive, a whole other league really. And wait till you hear the price...
English
NERO 7-DoF: Flexibility That Moves Like a Human
Redefine dexterity with NERO – perfect for lab setups and humanoid R&D.
✅ ±0.1mm Repeatability
✅ Multi-angle Mounting (upright/inverted/side)
✅ 4.8kg Ultra-lightweight Mobility
✅ 3.0kg Payload
✅ 580mm Working Radius
English
Prepare for the comeback of kinesthetic teaching!
Ilir Aliu@IlirAliu_
Robot policies fail on the hard parts of manipulation. The moment contact, friction, or force uncertainty shows up, the success rate drops fast. CR DAgger shows a very different path. You take a pre trained policy. You let a human correct it in the real world for a short time. And the system uses those corrections to learn a force aware residual policy that fixes the failure cases. The surprising part is the sample count❗️ They report improvements to near perfect performance on contact heavy tasks with only 50 to 100 episodes of correction. Key ideas in simple terms: ✓ A compliant interface so humans can give tiny corrections without stopping the robot ✓ A residual policy that adds the missing force reasoning to a position only base policy ✓ A fast update loop that makes the policy better with every bit of feedback They tested tasks like book flipping and belt assembly. The gains are large. It beats finetuning and it beats retraining from scratch. Thanks for sharing, @YifanHou2!! 📍Paper: arxiv.org/abs/2506.16685 Extended version with more experiments: compliant-residual-dagger.github.io/files/CR_DAgge… Code: github.com/yifan-hou/cr-d… —- Weekly robotics and AI insights. Subscribe free: scalingdeep.tech
English