MIT's New AI System Creates Its Own Robots

The system is inspired by the animal kingdom.
Chris Young
The photo credit line may appear like thisMIT

A new MIT system allows robot creators to simulate different robotic forms to determine which one will work best for their desired outcome.

The system, called RoboGrammar, simulates different robotic forms after developers key in the parts they want to use for their robot as well as the type of terrain their robot will need to navigate.

RELATED: NEW ROBOTS OFFER COVID-SAFE SOLUTIONS FOR LIVE EVENTS

Automating robot design

"Robot design is still a very manual process," Allan Zhao, the paper’s lead author and a PhD student in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), explains in a press release.

Zhao describes RoboGrammar as "a way to come up with new, more inventive robot designs that could potentially be more effective."

The team at MIT says that the primary inspiration for RoboGrammar came from the animal kingdom rather than from other robot designs — arthropods such as spiders and lobsters in particular.

MIT's New AI System Creates Its Own Robots
Several robot designs generated using RoboGrammar, Source: MIT

RoboGrammar operates in three sequential steps: defining the problem, drawing up possible robotic solutions, then selecting the optimal models.

Using these steps, RoboGrammar uses the rules of the graph grammar set up by the MIT team to design hundreds of thousands of potential robot structures. 

"It was pretty inspiring for us to see the variety of designs," says Zhao. "It definitely shows the expressiveness of the grammar."

Designing the optimal robot

In order to then evaluate which of the designs would work best, the team developed a controller for each robot with an algorithm called Model Predictive Control, which prioritizes rapid forward movement.

"The shape and the controller of the robot are deeply intertwined," says Zhao, "which is why we have to optimize a controller for every given robot individually."

Once each simulated robot is able to move, the researchers then seek the high-performing robots using a "graph heuristic search." This neural network algorithm iteratively samples and evaluates sets of robots, the researchers say. Ultimately, it allows the RoboGrammar program to select the optimal robot design for any given scenario.

All of this happens before the human designer has picked up a single tool or built a single prototype.

A program that essentially uses artificial intelligence to design robots will draw inevitable Skynet comparisons, but it could also help roboticists to build the best possible models for disaster response and other similar capabilities.

IELogoIELogo

Subscribe today

For full access to all features
and product updates.

%30 Save Quarterly

$25

$17.97

Quarterly

Subscribe Now
You can cancel anytime.
View Other Options

Already have an account? Log in

0 Comment
Already have an account? Log in