# MIT Technology Review profiles Danijar Hafner’s work on robots that plan ahead

[Canonical reading](https://escapemeridian.com/readings/mit-technology-review-profiles-danijar-hafner-s-work-on-robots-c7a91b0c)

MIT Technology Review’s profile, published September 8, 2026, describes Danijar Hafner’s work on AI agents that learn to handle unfamiliar situations. Hafner says his new startup continues his work on enabling AI to navigate environments it did not encounter during training, with humanoid robots as its physical embodiment. The profile explains that Hafner uses model-based reinforcement learning: he trains agents in world models, which emulate physical reality, so they can predict possible outcomes and plan actions. It says this approach aims to let robots respond to unexpected situations without the real-world trial and error traditionally used in robotics. The profile traces earlier tests in video games and describes DayDreamer as a step toward physical environments, where robots operated in novel settings and reacted to experiences such as being pushed over without specific training. It does not give measures or comparisons for those results, so their relevance to robots working reliably in people’s homes remains unclear. Hafner left Google DeepMind to form the startup in the fall of 2025; the profile does not disclose its name or much about its plans. He says he was interested in solving a problem “that would change the world.”

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

MIT Technology Review · Article

Original source date: Sep 8, 2026 (2026-09-08T10:34:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-19T19:28:12.586Z. First observed: 2026-11-19T18:34:00.985Z. Retrieved: 2026-11-19T18:35:59.263Z.

- [This AI entrepreneur is developing agents that can plan ahead for the unexpected](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Limits

The profile gives no measures or comparisons for the described results, so it does not establish how reliably the approach works in physical environments such as homes. The startup’s name and further plans are not disclosed in the retrieved text. Text as retrieved on November 19, 2026; originally published September 8, 2026.


## Open question

What evidence will show whether Hafner’s approach lets robots handle unfamiliar situations reliably in physical environments such as people’s homes?


## Source-stated

Hafner describes his new startup as continuing his work on enabling AI to navigate environments it did not encounter during training.

> he describes it as a continuation of his longtime work to enable AI to navigate environments it has not encountered in training.

- [Supporting source](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Source-stated

The profile says Hafner develops world models that emulate physical reality and trains agents within them to predict future outcomes.

> He develops world models—AI models designed to emulate physical reality—and trains agents within them.

- [Supporting source](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Source-stated

The profile says DayDreamer let robots operate in novel environments and react to new experiences without specific training.

> His DayDreamer project used the Dreamer algorithm to let robots operate themselves in novel environments and react to new experiences (such as being pushed over) without any specific training.

- [Supporting source](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Source-stated

Hafner says he was interested in solving a problem “that would change the world.”

> “I was interested in solving a problem,” he hints, “that would change the world.”

- [Supporting source](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Source and contributor context

MIT Technology Review · 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.


## AI agent

An AI system that chooses and carries out actions toward a goal, often using software tools.

- [Definition reference](https://www.ibm.com/think/topics/ai-agents)

## Reinforcement learning

Training by trial and error, in which a system is rewarded for the outcomes its designers want.

- [Definition reference](https://developers.google.com/machine-learning/glossary#reinforcement-learning-rl)

## World model

An AI system's internal model of how an environment behaves, used to predict what happens next or to plan.

- [Definition reference](https://arxiv.org/abs/1803.10122)

## Google DeepMind

Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.

- [Explore this context](/entities/google-deepmind)
- [Reference](https://deepmind.google/blog/)
- [Reference](https://deepmind.google/science/alphafold/)

## The World

The World is a public radio news program publishing international reporting on politics, society, labor and technology.

- [Explore this context](/entities/the-world)
- [Reference](https://theworld.org/)

## Editorial provenance

Version 1. Publication ID: ec79be54-b8cb-c8fd-2bbc-48b7fea8b81c.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-19T19:23:28.497Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical.

Observation: 4d03c240e45f15e55955fd1ebf59de74f491d727e099a6de10c4581c7de9e08f. Source text hash: 9e977496af7cbf30aea0df416beeffdb78594fbc0618ee14b81f4b523720dd18.


## Revision and source-check history

### Version 1 — 2026-11-19T19:28:12.586Z

MIT Technology Review’s profile, published September 8, 2026, describes Danijar Hafner’s work on AI agents that learn to handle unfamiliar situations. Hafner says his new startup continues his work on enabling AI to navigate environments it did not encounter during training, with humanoid robots as its physical embodiment. The profile explains that Hafner uses model-based reinforcement learning: he trains agents in world models, which emulate physical reality, so they can predict possible outcomes and plan actions. It says this approach aims to let robots respond to unexpected situations without the real-world trial and error traditionally used in robotics. The profile traces earlier tests in video games and describes DayDreamer as a step toward physical environments, where robots operated in novel settings and reacted to experiences such as being pushed over without specific training. It does not give measures or comparisons for those results, so their relevance to robots working reliably in people’s homes remains unclear. Hafner left Google DeepMind to form the startup in the fall of 2025; the profile does not disclose its name or much about its plans. He says he was interested in solving a problem “that would change the world.”

## Original source

MIT Technology Review · Article

Original source date: Sep 8, 2026 (2026-09-08T10:34:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-19T19:28:12.586Z. First observed: 2026-11-19T18:34:00.985Z. Retrieved: 2026-11-19T18:35:59.263Z.

- [This AI entrepreneur is developing agents that can plan ahead for the unexpected](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Limits

The profile gives no measures or comparisons for the described results, so it does not establish how reliably the approach works in physical environments such as homes. The startup’s name and further plans are not disclosed in the retrieved text. Text as retrieved on November 19, 2026; originally published September 8, 2026.


## Open question

What evidence will show whether Hafner’s approach lets robots handle unfamiliar situations reliably in physical environments such as people’s homes?


## Source-stated

Hafner describes his new startup as continuing his work on enabling AI to navigate environments it did not encounter during training.

> he describes it as a continuation of his longtime work to enable AI to navigate environments it has not encountered in training.

- [Supporting source](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Source-stated

The profile says Hafner develops world models that emulate physical reality and trains agents within them to predict future outcomes.

> He develops world models—AI models designed to emulate physical reality—and trains agents within them.

- [Supporting source](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Source-stated

The profile says DayDreamer let robots operate in novel environments and react to new experiences without specific training.

> His DayDreamer project used the Dreamer algorithm to let robots operate themselves in novel environments and react to new experiences (such as being pushed over) without any specific training.

- [Supporting source](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Source-stated

Hafner says he was interested in solving a problem “that would change the world.”

> “I was interested in solving a problem,” he hints, “that would change the world.”

- [Supporting source](https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/)

## Source and contributor context

MIT Technology Review · 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.


## AI agent

An AI system that chooses and carries out actions toward a goal, often using software tools.

- [Definition reference](https://www.ibm.com/think/topics/ai-agents)

## Reinforcement learning

Training by trial and error, in which a system is rewarded for the outcomes its designers want.

- [Definition reference](https://developers.google.com/machine-learning/glossary#reinforcement-learning-rl)

## World model

An AI system's internal model of how an environment behaves, used to predict what happens next or to plan.

- [Definition reference](https://arxiv.org/abs/1803.10122)

## Google DeepMind

Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.

- [Explore this context](/entities/google-deepmind)
- [Reference](https://deepmind.google/blog/)
- [Reference](https://deepmind.google/science/alphafold/)

## The World

The World is a public radio news program publishing international reporting on politics, society, labor and technology.

- [Explore this context](/entities/the-world)
- [Reference](https://theworld.org/)

## Editorial provenance

Version 1. Publication ID: ec79be54-b8cb-c8fd-2bbc-48b7fea8b81c.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-19T19:23:28.497Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical.

Observation: 4d03c240e45f15e55955fd1ebf59de74f491d727e099a6de10c4581c7de9e08f. Source text hash: 9e977496af7cbf30aea0df416beeffdb78594fbc0618ee14b81f4b523720dd18.


### uncertain — 2026-11-19T19:23:28.497Z — anthropic/claude-opus-5.5

Publication: ec79be54-b8cb-c8fd-2bbc-48b7fea8b81c

attribution — pass: The reading attributes the profile to MIT Technology Review and keeps Hafner's statements and Lillicrap-free descriptions attributed to the profile or Hafner. The page is a single feature article, not a listing; related-story links at the end are not used.

> he describes it as a continuation of his longtime work to enable AI to navigate environments it has not encountered in training.

summary — pass: The reading leads with the actor and his argument and explains model-based reinforcement learning and world models in plain language (also defined in the term library). It conveys significance (robots entering homes facing unseen layouts) and the gap in evidence. The question follows from the source. Omitting the Dreamer milestones is acceptable compression.

> if you want to send a robot into a person’s home, for example, it needs to be able to handle a floor plan and furniture it’s never seen before.

date — uncertain: September 8, 2026 matches the UTC day of originalPublished, with no hindsight. However, the page was modified on 2026-09-10 and carries an update note saying timeline information was added. The reading's statement that Hafner left Google DeepMind in the fall of 2025 may come from that later edit rather than the original piece, so its origin is unclear.

> Update: This story was updated to add timeline information.

support — pass: Claims match the source: startup as continuation of his work, world models emulating physical reality, agents making predictions, avoiding real-world trial-and-error, DayDreamer reacting to being pushed over, and the change-the-world quote. The limitation about absent measures is a fair, labeled gap. The fall 2025 timing concern is addressed under date.

> His DayDreamer project used the Dreamer algorithm to let robots operate themselves in novel environments and react to new experiences (such as being pushed over) without any specific training.

media — pass: No media was selected and no verified media candidates were supplied, so there is no media identity or relevance issue.

## Recorded references

Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter.

No public references recorded here yet.

[Reference data (JSON)](/readings/c7a91b0c360dda93cbb4fcd89cab0299f4abe4d06270301bfa2d180c79ec88b4/references.json)

## Corrections and challenges

[Send a correction or an opposing source](https://escapemeridian.com/corrections/)

[JSON record](https://escapemeridian.com/readings/mit-technology-review-profiles-danijar-hafner-s-work-on-robots-c7a91b0c.json)
