Skip to Content
Artificial intelligence

Why AI must learn to dress itself

November 6, 2018

An AI algorithm has finally conquered a task most people take for granted: putting on clothes in the morning.

Clumsy act: Getting dressed is surprisingly hard for a machine to figure out, because it involves wrangling a piece of very flexible material. Think about the awkward motions you go through when  donning a sweater: you have to tug the cloth and move your body in just the right way to get it over your arms and head without tearing it.

(Don’t) let it rip: Researchers at the Georgia Institute of Technology programmed a humanoid character to figure out the task by itself, even when the starting position and shape of the garment changed. They did so with a reinforcement-learning (RL) algorithm, a machine-learning technique. Inspired by the way we train animals, RL uses rewards and penalties to get an AI agent to achieve a desired goal. In this instance, the algorithm rewarded behaviors that led the humanoid to put its limbs and head through the right holes and penalized behaviors that could cause the garment to rip.

Chunk it up: Rather than program the motion of dressing as one long task with a single goal, the researchers broke it down into subtasks, such as grasping the front layer of a shirt, tucking a hand into the shirt’s opening, and pushing it through the sleeve. Each subtask required hours of simulation and optimization, but it made the final performance better able to cope with variations in the clothing.

Into the future: The ability to simulate complicated motor skills is relevant for computer animation and entertainment. It could also be used to advance robotics far later down the line by giving machines the ability to adapt to changing environmental conditions.

Deep Dive

Artificial intelligence

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

Google DeepMind’s new generative model makes Super Mario–like games from scratch

Genie learns how to control games by watching hours and hours of video. It could help train next-gen robots too.

What’s next for generative video

OpenAI's Sora has raised the bar for AI moviemaking. Here are four things to bear in mind as we wrap our heads around what's coming.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at with a list of newsletters you’d like to receive.