UC Berkeley's Ken Goldberg argues 10,000 hours of open-source data is insufficient to train robot foundation models · Digg

/Tech8h ago

UC Berkeley's Ken Goldberg argues 10,000 hours of open-source data is insufficient to train robot foundation models

Story Overview

Aggregated open-source robotics datasets have hit roughly 10,000 hours across efforts like AgiBotWorld, Open X-Embodiment, and ABC, prompting claims that this scale now lets anyone train a working robot foundation model. Ken Goldberg at UC Berkeley pushes back, pointing to the practical headaches of blending recordings from robots with mismatched arms, grippers, and sensors.

161991720136.6K

Original post

Ken Goldberg@Ken_Goldberg#508inTech

Chris, 10k hours (approx 1 year) is roughly 0.001 % of internet scale data. Plus it’s not clear how to merge data from different embodiments.

Chris Paxton@chris_j_paxton

I think this means you can collect ~10k hours just from open source datasets, which means basically anyone should be able to build a decent robot foundation model:

500 from the new BitRobot dataset 500 from galaxea https://huggingface.co/datasets/OpenGalaxea/Galaxea-Open-World-Dataset 3000 from agibot https://huggingface.co/datasets/agibot-world/AgiBotWorld2026 ~3000 from Open X embodiment (though it's mostly pretty bad data) https://robotics-transformer-x.github.io/ 830 from EgoDex https://github.com/apple/ml-egodex ~30 from humanoid-everyday (but good quality) https://huggingface.co/datasets/USC-PSI-Lab/humanoid-everyday 3500 from ABC https://huggingface.co/datasets/XDOF/ABC-130k

2:04 AM · Jun 25, 2026 · 7.7K Views

Open Question

Merging hardware quirks remains the sticking point

Goldberg’s objection centers on embodiment differences rather than raw volume alone, yet no detailed benchmarks or failure cases from his side have surfaced so far.

Developer Impact

Next moves hinge on cross-platform fixes

Without clearer evidence on how much extra data or new techniques would close the gap, teams eyeing these datasets for foundation models face an unresolved transfer hurdle.

Sentiment

Positive users highlight accelerating cross-embodiment research and potential of the 10k-hour open-source robot datasets for foundation models, while negative users argue the scale remains insufficient.

Pos

75.0%

Neg

25.0%

4 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Related links

GitHub - apple/ml-egodex: EgoDex: Learning Dexterous Manipulation from Large-Scale Egocentric Video

GITHUBVia

agibot-world/AgiBotWorld2026 · Datasets at Hugging Face

Posts from X

Most Activity

VIEWS7.5KBOOKMARKS25LIKES15

Vai Viswanathan@vai_viswanathan

@chris_j_paxton check out http://datasets.bot! Been aggregating these!

12h7.5K1525

RETWEETS12

Chris Paxton@chris_j_paxton

I think this means you can collect ~10k hours just from open source datasets, which means basically anyone should be able to build a decent robot foundation model:

500 from the new BitRobot dataset 500 from galaxea https://huggingface.co/datasets/OpenGalaxea/Galaxea-Open-World-Dataset 3000 from agibot https://huggingface.co/datasets/agibot-world/AgiBotWorld2026 ~3000 from Open X embodiment (though it's mostly pretty bad data) https://robotics-transformer-x.github.io/ 830 from EgoDex https://github.com/apple/ml-egodex ~30 from humanoid-everyday (but good quality) https://huggingface.co/datasets/USC-PSI-Lab/humanoid-everyday 3500 from ABC https://huggingface.co/datasets/XDOF/ABC-130k

Vai Viswanathan@vai_viswanathan

500 hours of data 🤯

We’re still far from internet scale, but incredible that researchers are releasing open source data

12h25.5K128178

REPLIES1

Vai Viswanathan@vai_viswanathan

@Ken_Goldberg Any thoughts on promising directions for cross-embodiment transfer?

I’ve been looking into Latent Action Models, and it seems like there’s a lot of room for research in that space.

6h1111

Chris Paxton@chris_j_paxton

Ah another 1k from MolmoAct2, how could I forget: https://huggingface.co/collections/allenai/molmoact2-datasets

Some overlap with Open X though

12h1.2K51

Chris Paxton@chris_j_paxton

Another 3k and we're getting close to 15k total data: https://www.modelscope.cn/datasets/Eureka1064/FTP-1-Dataset

12h88332

Rob@roblee_rl

@chris_j_paxton If we include UMI-style data, there's 13,000 hours from https://huggingface.co/datasets/genrobot2025/10Kh-RealOmin-OpenData and 8000 hours from https://yubi.airoa.io/ (releasing soon apparently).

11h22442

Nick K@nicks_robots

@chris_j_paxton How are you determining which datasets are good/bad quality chris?

12h2811

Chris Paxton@chris_j_paxton

@nicks_robots Vibes, and how often people don't use it despite it's availability in the case of open x

12h2645

Chetan@chetan_

If you include UMI style datasets there is even more.

iirc pi0 was trained on ~10k hours so definitely sizable enough to do some interesting work!

Chris Paxton@chris_j_paxton

I think this means you can collect ~10k hours just from open source datasets, which means basically anyone should be able to build a decent robot foundation model:

500 from the new BitRobot dataset 500 from galaxea https://huggingface.co/datasets/OpenGalaxea/Galaxea-Open-World-Dataset 3000 from agibot https://huggingface.co/datasets/agibot-world/AgiBotWorld2026 ~3000 from Open X embodiment (though it's mostly pretty bad data) https://robotics-transformer-x.github.io/ 830 from EgoDex https://github.com/apple/ml-egodex ~30 from humanoid-everyday (but good quality) https://huggingface.co/datasets/USC-PSI-Lab/humanoid-everyday 3500 from ABC https://huggingface.co/datasets/XDOF/ABC-130k

3h1.5K114

Chris Paxton@chris_j_paxton

@vai_viswanathan Ah nice thank you

12h5741

Sammo@sammo3x

@chris_j_paxton In the grand scheme of things its basically nothing I think we need more than that

9h95

zeke@zshobbs

@chris_j_paxton i think i need to expand my nas!!!!!

7h94

zeke@zshobbs

@Ken_Goldberg Look at the GR00T N1 paper from Nvidia they trained on lots of different embodiments.

https://arxiv.org/abs/2503.14734

5h52

clankr@clankrmedia

@Ken_Goldberg Definitely. A grounded engine should convert data for cross-embodiment. If this problem is solved, 10k hours of data is enough for a foundation model. Interesting work is happening there; it is better to watch them.

4h31

Michael Cho - Rbt/Acc@micoolcho

@Ken_Goldberg Yes teleop data is tough to collect; expect more to come but grow only at linear scale.

Human data (eg. ego videos, wrist cameras, tactile gloves, etc) can scale way more though.

3h22

Скандинавское математическое нытье 🍀@V4ldeLund

@vai_viswanathan @Ken_Goldberg I think cosmos just straight away threw away real actions, trained on pseudo delta actions and then post train on Droid

5h21

Justin Chen@ch3njus

@Ken_Goldberg Cross embodiment research generally seems to be picking up pace. Recent work from Karen Liu on masking out grippers in egocentric to retarget seems promising and recent humanoid wb retargeting seem pretty decent on demos.

4h13

fourier@fourierCO

@Ken_Goldberg guess what? robotics policies don't need internet scale data to train

4m3

Taro Bushidō@techietaro

@Ken_Goldberg Exactly. Rather than merging raw trajectories across embodiments, train each within its own world model first. Then combine generalizable features at a higher abstraction layer.

7h1