Rodney Brooks · Sep 27, 2025
Rodney Brooks argues humanoid robots lack the touch and dexterity their promise requires
Markdown · JSON · Original source
Rodney Brooks’s post was published September 27, 2025. Brooks argues that humanoid robots will not learn human-like dexterity from videos alone: human manipulation depends on touch and force information that visual recordings leave out. He contrasts this with speech and image systems, which, he argues, rely on carefully engineered ways of preparing their inputs. The problem, in his account, is not simply a need for more data; systems must collect the right data and learn how sensing relates to plans for action. Brooks says some academic experiments go beyond visual data, but have not yet demonstrated real dexterity. He also argues that full-sized, two-legged humanoid robots are dangerous to be near when they fall, making safety consequential if people are expected to share spaces with them. His conclusion is that humanoid robots will take many forms for specialized jobs and retain the name, while the designs he discusses will be gone and money will have been spent trying to improve their performance. How researchers can collect touch and force data and connect it to plans for action well enough to demonstrate dexterity remains unknown.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
Rodney Brooks · Article
Source details
Original source date: Sep 27, 2025 (2025-09-27T00:02:56.000Z).
Date basis: Date supplied by the publisher’s feed.
Added to Escape Meridian: 2026-11-12T00:58:23.038Z. First observed: 2026-10-23T19:23:44.850Z. Retrieved: 2026-10-23T19:25:38.311Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 23, 2026; originally published September 27, 2025. Brooks says he is not presenting a comprehensive case as one would expect in a formal peer-reviewed academic research paper. The article’s correction to its description of the MIT glove-and-robot-hand experiment says the person receives joint-level force feedback, not active stimulation of the fingertips and palm.
Open question
How will researchers collect touch and force data, and connect it to plans for action, well enough to demonstrate dexterity?
Source-stated
Brooks argues that learning dexterity from video alone misses information about touch and force that human manipulation uses.
Collecting just visual data is not collecting the right data. There is so much more going into human dexterity that visual data completely leaves out.
Source-stated
Brooks argues that successful end-to-end learning systems rely on domain-specific preparation of their data, and that dexterity learning needs appropriate data and learning targets.
If any one or any group is to succeed they will likely have to collect the both the right data, and learn the right thing.
Source-stated
Brooks says some academic experiments go beyond visual data, but have not yet demonstrated real dexterity.
There are some exciting and promising experiments going on in academic laboratories, but they have not yet gotten close to demonstrating any real dexterity.
Source-stated
Brooks argues that full-sized, two-legged humanoid robots are not currently safe for people to be near, and says falls can injure people.
The trouble is that human sized two legged walking humanoid robots are not currently safe for humans to be around.
Source-stated
Brooks predicts that humanoid robots will take many forms and perform specialized jobs, while still being called humanoid robots.
There will be many, many robots with different forms for different specialized jobs that humans can do. But they will all still be called humanoid robots.
Source-stated
Brooks predicts that today’s humanoid robots will be gone and that money will have been spent trying to improve their performance.
And a lot of money will have disappeared, spent on trying to squeeze performance, any performance, from today’s humanoid robots. But those robots will be long gone and mostly conveniently forgotten.
Source and contributor context
Rodney Brooks · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Editorial provenance
Version 1. Publication ID: 5a08c821-aa6c-c025-e966-1713e675f491.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T00:53:42.291Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: 56a89566c8ee464cd17d2fb1822f49a083513c3ef3e5722e647589086d6b7b34. Source text hash: bb1c8786205c51d79d83f91e3b1eeca72704bea0eec8a0e7e603df8d98bd3441.
Revision and source-check history
Version 1 · 2026-11-12T00:58:23.038Z
Rodney Brooks argues humanoid robots lack the touch and dexterity their promise requires
Rodney Brooks’s post was published September 27, 2025. Brooks argues that humanoid robots will not learn human-like dexterity from videos alone: human manipulation depends on touch and force information that visual recordings leave out. He contrasts this with speech and image systems, which, he argues, rely on carefully engineered ways of preparing their inputs. The problem, in his account, is not simply a need for more data; systems must collect the right data and learn how sensing relates to plans for action. Brooks says some academic experiments go beyond visual data, but have not yet demonstrated real dexterity. He also argues that full-sized, two-legged humanoid robots are dangerous to be near when they fall, making safety consequential if people are expected to share spaces with them. His conclusion is that humanoid robots will take many forms for specialized jobs and retain the name, while the designs he discusses will be gone and money will have been spent trying to improve their performance. How researchers can collect touch and force data and connect it to plans for action well enough to demonstrate dexterity remains unknown.
Original source
Rodney Brooks · Article
Source details
Original source date: Sep 27, 2025 (2025-09-27T00:02:56.000Z).
Date basis: Date supplied by the publisher’s feed.
Added to Escape Meridian: 2026-11-12T00:58:23.038Z. First observed: 2026-10-23T19:23:44.850Z. Retrieved: 2026-10-23T19:25:38.311Z.
Limits
Text as retrieved on October 23, 2026; originally published September 27, 2025. Brooks says he is not presenting a comprehensive case as one would expect in a formal peer-reviewed academic research paper. The article’s correction to its description of the MIT glove-and-robot-hand experiment says the person receives joint-level force feedback, not active stimulation of the fingertips and palm.
Open question
How will researchers collect touch and force data, and connect it to plans for action, well enough to demonstrate dexterity?
Source-stated
Brooks argues that learning dexterity from video alone misses information about touch and force that human manipulation uses.
Collecting just visual data is not collecting the right data. There is so much more going into human dexterity that visual data completely leaves out.
Source-stated
Brooks argues that successful end-to-end learning systems rely on domain-specific preparation of their data, and that dexterity learning needs appropriate data and learning targets.
If any one or any group is to succeed they will likely have to collect the both the right data, and learn the right thing.
Source-stated
Brooks says some academic experiments go beyond visual data, but have not yet demonstrated real dexterity.
There are some exciting and promising experiments going on in academic laboratories, but they have not yet gotten close to demonstrating any real dexterity.
Source-stated
Brooks argues that full-sized, two-legged humanoid robots are not currently safe for people to be near, and says falls can injure people.
The trouble is that human sized two legged walking humanoid robots are not currently safe for humans to be around.
Source-stated
Brooks predicts that humanoid robots will take many forms and perform specialized jobs, while still being called humanoid robots.
There will be many, many robots with different forms for different specialized jobs that humans can do. But they will all still be called humanoid robots.
Source-stated
Brooks predicts that today’s humanoid robots will be gone and that money will have been spent trying to improve their performance.
And a lot of money will have disappeared, spent on trying to squeeze performance, any performance, from today’s humanoid robots. But those robots will be long gone and mostly conveniently forgotten.
Source and contributor context
Rodney Brooks · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Editorial provenance
Version 1. Publication ID: 5a08c821-aa6c-c025-e966-1713e675f491.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T00:53:42.291Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: 56a89566c8ee464cd17d2fb1822f49a083513c3ef3e5722e647589086d6b7b34. Source text hash: bb1c8786205c51d79d83f91e3b1eeca72704bea0eec8a0e7e603df8d98bd3441.
uncertain · 2026-11-12T00:53:42.291Z · anthropic/claude-opus-5.5
Publication: 5a08c821-aa6c-c025-e966-1713e675f491
attribution — pass: This is a single essay on Rodney Brooks's own site, written in his first-person voice. The reading consistently presents the arguments, predictions and safety claims as his rather than as established fact.
In this post I explain why today’s humanoid robots will not learn how to be dexterous
summary — uncertain: The core argument is faithful and understandable: touch and force are missing from video training, earlier successes relied on engineered input stages, and walking is unsafe. However, the final sentence presents Brooks's forecast without its stated scope. He frames it as the next fifteen years. Smallest repair: say his conclusion concerns the next fifteen years. Minor point: speech and image systems are a comparison he uses to support his argument, not strictly a contrast.
That is the next fifteen years for you.
date — pass: The originalPublished instant 2025-09-27T00:02:56Z falls on the UTC day September 27, 2025, which the reading states. The evidence kind is a publisher page, so no archive disclosure is needed. Forecasts are reported as Brooks's predictions, and no later events or hindsight framing are used.
support — pass: Each claim is source-stated with an exact excerpt. The limitations accurately reflect Brooks's disclaimer about not offering a peer-reviewed case and his own correction about the MIT glove experiment. No unsupported benefits, costs or stakeholder claims are added.
There are some exciting and promising experiments going on in academic laboratories, but they have not yet gotten close to demonstrating any real dexterity.
media — pass: No media is selected, and no verified candidates were supplied, so there is no identity or relevance problem.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.