Imagine having a continuum soft robotic arm bend around a bunch of grapes or broccoli, adjusting its grip in real time as it lifts the object. Unlike traditional rigid robots that generally aim to avoid contact with the environment as much as possible and stay far away from humans for safety reasons, this arm senses subtle forces, stretching and flexing in ways that mimic more of the compliance of a human hand. Its every motion is calculated to avoid excessive force while achieving the task efficiently.

In the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and Laboratory for Information and Decisions Systems (LIDS) labs, these seemingly simple movements are the culmination of complex mathematics, careful engineering, and a vision for robots that can safely interact with humans and delicate objects.

Soft robots, with their deformable bodies, promise a future where machines move more seamlessly alongside people, assist in caregiving, or handle delicate items in industrial settings. Yet that very flexibility makes them difficult to control. Small bends or twists can produce unpredictable forces, raising the risk of damage or injury. This motivates the need for safe control strategies for soft robots.

When it comes to training robots to perform agile, single-task motor skills, such as handstands or backflips, artificial intelligence methods can be very useful. But if you want to train your robot to perform multiple tasks—say, performing a backward flip into a handstand—things get a little more complicated.

“We often want to train our robots to learn new skills by compounding existing skills with one another,” said Ian Abraham, assistant professor of mechanical engineering. “Unfortunately, AI models trained to allow robots to perform complex skills across many tasks tend to have worse performance than training on an individual task.”

To solve for that, Abraham’s lab is using techniques from optimal control—that is, taking a mathematical approach to help robots perform movements in the most efficient and optimal way possible. In particular, they’re employing hybrid control theory, which involves deciding when an autonomous system should switch between control modes to solve a task. The research is published on the arXiv preprint server.

Contribute to deepseek-ai/DeepSeek-Math-V2 development by creating an account on GitHub.

AlphaEvolve, an AI system created by Google DeepMind, is helping mathematicians do research at a scale that was previously impossible — even if it does occasionally “cheat” to find a solution

Large language models (LLMs) like ChatGPT can write an essay or plan a menu almost instantly. But until recently, it was also easy to stump them. The models, which rely on language patterns to respond to users’ queries, often failed at math problems and were not good at complex reasoning. Suddenly, however, they’ve gotten a lot better at these things.

A new generation of LLMs known as reasoning models are being trained to solve complex problems. Like humans, they need some time to think through problems like these—and remarkably, scientists at MIT’s McGovern Institute for Brain Research have found that the kinds of problems that require the most processing from reasoning models are the very same problems that people need to take their time with.

In other words, they report in the journal PNAS, the “cost of thinking” for a reasoning model is similar to the cost of thinking for a human.