← All research storiesEngineering

PreprintEngineering

Tell a robot “don’t spill on my laptop,” and this AI helps it figure out you mean all electronics

MIT and NYU researchers used a large language model to help a simulated robot arm turn one-off corrections into common-sense rules, like steering clear of anything fragile.

How far alongPreprintA manuscript shared publicly, often before or alongside journal review. Check its separate journal and review status (each story says what we know). Results may change.
How soon could it reach you?A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
Where it comes fromarXiv (preprint server); to be presented at the Conference on Robot LearningMIT; New York University · Nov 9, 2026Read the original →

What the scientists did

Robots plan their movements with math, keeping a list of things in the room that matter for the job. Today, if you tell a robot “don’t drop that on my mug,” it adds the mug to its avoid list, and that’s all. You’d have to name every fragile object one by one, and the list keeps growing even after things leave the room.

The team built FLEX, a three-step system. “Add” puts the object you mentioned on the robot’s list. “Generalize” asks a large language model (GPT-4o) to look at everything on the list and guess the bigger rule, such as “avoid fragile things.” “Mask” uses the robot’s camera system to drop objects that are no longer there. The whole update takes about five seconds. Volunteers tested it by typing commands to a simulated robot arm doing three tasks: serving coffee, moving a candle and carrying a bowling ball.

Words to know

Motion planning

How a robot works out a path to move from one place to another without hitting things.

  • A robot arm carrying a cup across a messy table
  • A robot vacuum finding its way around chairs

Feature (in robotics)

Something in the environment the robot pays attention to when planning its moves.

  • A laptop to avoid
  • The edge of a table
  • A stack of paper

Generalization

Taking a lesson from one example and applying it to similar cases.

  • From “don’t spill on my laptop” to “don’t spill on any electronics”
  • From “stay away from the paper” to “stay away from anything that can burn”

What they found

In all three simulated tasks, the robot followed each command literally and also applied a broader rule: it avoided electronics (both the laptop and a nearby phone) after “don’t spill on my laptop,” stayed away from flammable objects after “stay away from the stack of paper,” and avoided fragile items after “don’t drop on the mug.”

The university news release does not report success rates or other test numbers, and we were not able to open the full paper to check them.

~5 secondstime for FLEX’s three steps to update the robot’s plan
3simulated tasks tested

Why it matters to you

If home robots ever help you cook, clean or care for a family member, you won’t want to program them with a rulebook. Systems like FLEX aim to let you speak normally, the way you would to a person, and have the robot pick up what you care about from a sentence or two.

Where this could lead

Easier-to-teach home and assistive robots

Robots that learn household preferences from casual instructions instead of careful programming.

3–10 years

Safer robots around people

Robots that infer “be careful with anything like this” after one warning, instead of making the same kind of mistake again.

3–10 years

These are possibilities the research points toward, not promises. Most early findings take years of testing before they reach everyday life, and some never do.

Why scientists care

Large language models have soaked up a lot of everyday common sense from text, such as the fact that liquids ruin devices. Researchers are trying to connect that knowledge to robots’ physical decisions, so machines can understand what people mean beyond their literal words.

Keep in mind

This is a preprint that has not yet been peer-reviewed, and all tests were in simulation, not with a real robot in a real home. A language model’s guesses can also be wrong or go too far, for example by avoiding things the user didn’t care about. We read the university news release; the source date shown is the release date because we could not open the arXiv page to confirm when the paper was posted.

How far along is the science?

  • Peer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
  • PreprintA manuscript shared publicly, often before or alongside journal review. Check its separate journal and review status (each story says what we know). Results may change.
  • Conference talkPresented to other scientists at a meeting. Usually short and early; a full paper may come later.
  • Agency reportPublished by a science agency such as NASA, NOAA or NIH. Reviewed internally, but not by an outside journal.
  • Press releaseAn announcement from a university or company. Useful, but always check the research it describes.

How soon could it reach you?

  • Could reach you soonCould show up in products, services or advice soon. Our rough editorial guess is about 1 to 3 years, not a forecast from the research.
  • A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
  • Long-term scienceBasic research that builds knowledge. Real-world uses, if any, are far off.
  • Small stepA small update to earlier work. Useful to researchers, but not a change you would notice yet.

Every story links to its original source: the paper itself, or the conference program when no paper is out yet. Explanations are written with AI, usually from the paper’s abstract (each story says what it was written from), then checked against the source. We only use journals, recognized preprint servers, science agencies, scientific conferences and established science news outlets.