Gleb Zarin (@zaringleb) header
Gleb Zarin (@zaringleb) profile picture

Robot Learning

London, UK
Joined February 2015
Agentic AI will finally democratize robotics. @GoogleDeepMind shipped the Antigravity robotics plugin, so we tried it. 2 prompts, ~20 min: Hydrogen set up our FR3 Duo, made it wave & packed LEGO in a box. Enthusiasts, deployment companies: you can see where this is going 👀
13
28
14
257
18,492
Gleb Zarin retweeted
Our team finished 9th in the 2026 BEHAVIOR Challenge and received the Open Source award 🏆 Huge thanks to my teammates @zaringleb, @atindranair, @keertimanney, @vkleban, @mirkokiefer and @ayoda1234 ! As you may guess from the prize name, we've open-sourced our solution: code, weights and a detailed tech report. Here is a summary 👇
1
4
1
29
1,502
The two mystery models are in 👀: It's Gemini 4 Carbon & Mistral Large 4! The Reality Check: Agentic is now complete. • 6 frontier models • 5,400 real-world rollouts • Spatial generalization measured Here are the results 👇 1/5
2
14
1
43
1,468
The age of agentic robotics is here, but most evals treat agents as tools instead of what they are: intelligent tool users. Like Homo habilis in human evolution, intelligent tool use unlocks a leap in performance. Today, we're releasing the largest systematic eval campaign. 1/12
18
21
14
137
406,660
Towards an agent-native CERN of AI robotics AI agents can now run the whole robotics research loop in hours: code, train, evaluate. But only in simulation. So we built a robot farm agents can control through an API. First test: can they replicate recent robotics papers? 1/17
8
14
3
70
3,685
Gleb Zarin retweeted
MolmoAct 2, our fully open robotics model, takes the top spot for overall task success on the independent Reality Check benchmark after task-specific fine-tuning. Thanks @Nicolas_Keller for putting MolmoAct 2 to the test and sharing your methods and results with the community. 🚀
The Era of Evals is coming to AI robotics. Today we’re launching Reality Check: our leaderboard & first public robot manipulation benchmark. 14,400 real-world rollouts. Four models. Multiple tasks & data regimes. Object placements. Confidence intervals. FR3 Duo stations. 1/10
2
12
2
78
13,604
The Era of Evals is coming to AI robotics. Today we’re launching Reality Check: our leaderboard & first public robot manipulation benchmark. 14,400 real-world rollouts. Four models. Multiple tasks & data regimes. Object placements. Confidence intervals. FR3 Duo stations. 1/10
60
71
34
692
120,460
Congrats to the @odysseyml team on Odyssey-3!
🤖 Made with AI
Today we’re unveiling Odyssey-3, a big step forward for foundation world models. It can control robots, power humanoids, drive cars (on the roads of India!), train AIs, pilot drones, and even play video games. We can’t wait to see what intelligent systems it enables.
1
5
25
1,688
Gleb Zarin retweeted
Our team won the BEHAVIOR-1K Challenge at @NeurIPSConf It started as a random call with @zaringleb (who I didn't know back then), who told me about the competition and invited me to join the team, and turned into two months of evenings and weekends teaching the robot to do home chores in simulation. I learned a lot and tried out a few new VLA-related ideas I'd had in my head for a while. We will share our solution: tech report, code, model weights, and a detailed video about the competition and our approach soon. I will also present a summary of the approach at the NeurIPS 2025 BEHAVIOR-1K workshop this Sunday (online). But right now, you can watch a robot autonomously completing a bunch of tasks controlled by our policy, from simple 30-second to multi-step 10+ minutes problems. This video is cherry-picked; our score is 26% (including partial successes), and we will share more details, including failed attempts, later.
4
3
20
1,253
Really cool guide to Robot Learning from one of @LeRobotHF community member (shared on our discord) robot-learning.notion.site/r…
4
63
376
21,616
Sorrow for Ukrainian people, fear for my friends and lack of any hope for a meaningful future in my country is all that was left by Putin's war.
The best answer to the question about an existential threat to humanity from AI is given by Steven Pinker. Absolutely agree with it, the question is a bs to look smart taking about. youtu.be/epQxfSp-rdU?t=668
1
1
One example, the number one cause of death worldwide is cardiovascular disease. It takes 17 million lives a year en.wikipedia.org/wiki/List_o… And it is known for decades that CDC is a preventable disease thelancet.com/journals/lance… Alas, it is not "cool" or "smart" to speak about it.