RoboColosseum (@RoboColosseum) header
RoboColosseum (@RoboColosseum) profile picture

Benchmarking what robot models can actually do in the real world. Free eval hours for academia ↓

Joined October 2026
Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)Photo by RoboColosseum (@RoboColosseum)
Introducing RoboColosseum An open robot eval initiative by @huggingface @BitRobotNetwork @eaic_sg @DJiafei Access: - 100+ robots (@UnitreeRobotics G1s, @LeRobotHF SO-101s, YAMs, Franka and more) - 140+ in-the-wild scenes - Open rollout data - 10,000 FREE eval hrs Thread ↓
55
73
85
458
138,449
1/ After a wave of model releases at @corl_conf, we gave five models a zero-shot task: fold a towel with a bimanual YAM. Some runs had latency issues, so timings aren’t apples to apples. Still, the demos give a first look at how each model handles the task. ↓
5
3
2
21
1,281
3/ Dream-1 made an initial fold but left the towel bunched. MolmoAct2 held it with both grippers; Gemini repositioned without moving it.
1
3
121
4/ CoRL was a just first look. Next, we’ll test these models across more tasks and real-world environments with RoboColosseum. Want to test your own policy? Academic labs and researchers can apply for a share of our 10,000 free eval hours: app.youform.com/forms/riubyc…
3
82
RoboColosseum retweeted
That’s why we evaluate robots at scale, in the wild, with @RoboColosseum. Physics just can't go wrong. We’re giving away 10,000 hours of free real-world robot evaluation. Sign up now: app.youform.com/forms/riubyc…
in our new blog, we take a look at RoboDojo sim, which are base on NVIDIA's Isaac Sim and it is a good robo benchmark used in Fable 5.2 model card. the video pretty much show what we found. the simulation is not even wrong. I'm now fully understand why robotics experts "hate" the past simulator. but from the first principle, I do believe LLM powered agent could do much much better simulation with the minimal but necessary real-world calibration is the right way we build embodied intelligence. real-world are slow, expensive, hard to scale, hard to get labelled, but sim not. then we partially solve the data problem for robotic and we could scaling up the compute!
4
25
2,275
RoboColosseum retweeted
Check out our keynote at #CoRL2026 on @RoboColosseum ! I shared our mission, vision and future of the project as a proving ground for physical intelligence!
Last week we took the keynote stage at @corl_conf. Our mission: more robots in the real world, evaluations at scale to show what policies can do today, and all of it open-sourced. Catch RoboColosseum's founding member @DJiafei’s talk ↓
1
4
28
1,409
Last week we took the keynote stage at @corl_conf. Our mission: more robots in the real world, evaluations at scale to show what policies can do today, and all of it open-sourced. Catch RoboColosseum's founding member @DJiafei’s talk ↓
1
2
2
20
3,791
RoboColosseum retweeted
Great insights and eval! Looking strong with MolmoAct2 @weikaih04
most VLAs lose the vision-language skills of their VLM during pre-training by isolating the VLM weights of open-source VLAs, we can evaluate them on vision-language benchmarks. three categories of models emerge
5
1
18
2,502
You can benchmark an LLM from a laptop, but robot policies need real hardware / environments for testing. Yesteday, @pepijn2233 shared how @huggingface is releasing every RoboColosseum eval rollout, including failures. What a great to wrap up @corl_conf!
2
8
1
30
3,716
RoboColosseum retweeted
Replying to @pepijn2233
@pepijn2233 from @LeRobotHF explains why scaling in the wild at @RoboColosseum is very important for robotics and pushing physical AI progress.
1
2
11
422
RoboColosseum retweeted
We had a great first at @corl_conf, kicking off with a main stage keynote from co-founder @micoolcho and a product demos from @GroundedSI @RoboSenseLiDAR @moxianmxkj We have a few more lightning talks today. Check out the schedule below!
5
7
1
36
4,170
“Fold the yellow towel.” Dream-1 from @dreamlabs_ai, running on a bimanual YAM at CoRL @corl_conf 🦾
9
1
32
3,018
RoboColosseum retweeted
We are already evaluating Fable 5.2 live for head to head comparison at @BitRobotNetwork booth for our @RoboColosseum !
2
4
49
2,331
RoboColosseum retweeted
Singapore is entering the open-source robotics arena 🦾 Great crowd for @ruijie_sg repping for @eaic_sg to chat about how the island-state is pushing forward open benchmarks w/ @RoboColosseum. We’re hosting lighting talks all day at @BitRobotNetwork CoRL (408). Swing by!
7
5
1
34
3,044
RoboColosseum retweeted
We're testing out the latest Fable release from @claudeai at @RoboColosseum @corl_conf It was able to grip the towel zero-shot with the YAM arms, so far!
Let's put the latest release of @claudeai Fable 5.2 to the test at @corl_conf Swing by the BitRobot Booth (408) to see Fable zero-shot towel-folding, pick/place tasks, or even your own prompt!
6
3
23
3,044
Let's put the latest release of @claudeai Fable 5.2 to the test at @corl_conf Swing by the BitRobot Booth (408) to see Fable zero-shot towel-folding, pick/place tasks, or even your own prompt!
Replying to @claudeai
Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do. On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.
5
4
2
41
5,910
RoboColosseum retweeted
How far can Fable 5.2 @claudeai go on tasks it's never seen before? 👀 We're bringing Fable 5.2 to the real world at the BitRobot booth with @Robocolosseum at @corl_conf! No cherry-picked demos. Come see for yourself! 🚀
Replying to @claudeai
Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do. On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.
1
2
8
2,711
RoboColosseum retweeted
We will be showcasing Fable 5.2 live head to head comparison today at #CoRL2026 BitRobot booth for @RoboColosseum ! If you are keen to find out its capabilities comparing to other models, please drop by the booth 408!
Replying to @claudeai
Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do. On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.
1
4
45
2,237
Exhibits open soon! Swing by to check out this YAM arm pick up a screwdriver and place it in a box in <15 seconds. Can you guess which policy was running? 👀 Come test different tasks, including your own, through blind evals on our set up at the @corl_conf @BitRobotNetwork!
1
3
1
16
1,560
1/ What does a more open robotics ecosystem look like? In their @corl_conf keynote, @micoolcho @DJiafei will share how open-source data, accessible hardware, and real-world evaluation can advance robotics research. Catch the session at the Hilton Austin on Nov 9, 4:40 PM.
3
5
1
21
1,358
2/ And there’s more at the @BitRobotNetwork booth 408! Hear from @RoboColosseum contributors during our ecosystem lightning talks and see live policy evaluations in action. Come say hi! xitter.cf/BitRobotNetwork/status…
2/ Day 1 | Nov 9th 13:00 RoboColosseum - @DJiafei @NUSComputing MAGIC Lab 13:30 MolmoAct2 - @DJiafei @hq_fang @allen_ai 15:00 RoboCommons - @UW @YiruHelenWang 15:30 RoboEval Results - @UW @allen_ai 16:00 RoboTTS @NVIDIA Zhenzhen Liu
1
4
188