Researchers at Northwestern University developed the first artificial intelligence to date that can intelligently design robots from scratch. To test it, they asked the AI to “design a robot that can walk across a flat surface.” A few seconds later, the AI had created this purple block — a successfully walking robot with legs and rear fins.

The AI isn’t just fast but also runs on a lightweight personal computer, creating novel structures from scratch. This contrasts with other AI systems that usually need energy-hungry computers and big datasets. And even after going through all that data, those systems are limited to the constraints of human creativity, only mimicking past works.

“We discovered a very fast AI-driven design algorithm that bypasses the traffic jams of evolution, without falling back on the bias of human designers,” Sam Kriegman, one of the researchers, said in a news release. “We told the AI that we wanted a robot that could walk across land. Then we simply pressed a button and presto!”

In 2020, Kriegman developed xenobots, the first living robots entirely made from biological cells. Now, he sees AI as the next advance in their quest to explore the potential of artificial life. The robot itself is modest but Kriegman believes it represents the first step in a new era of AI-designed tools that can act directly on the world.

The potential of AI

Although the new AI program has the ability to initiate from any given prompt, Kriegman and his research team’s starting point was a straightforward task: designing a tangible machine capable of terrestrial locomotion. After setting this objective, the researchers relinquished control, allowing the AI to autonomously handle the rest.

The computer started with a block the size of a bar of soap that could jiggle but not walk. Aware that it hadn’t met its goal, the AI quickly changed the design. Each attempt helped it to identify flaws and update the structure. Eventually, the AI created a robot that could walk half its body length per second — half the speed of a human stride.

Surprisingly, the design process of the robot, from a shapeless block to a walking robot, took just 26 seconds on a laptop. “Now anyone can watch evolution in action as AI generates better and better robot bodies in real time,” Kriegman said in a news release. “Evolving robots previously required weeks of trial and error on a supercomputer.”

The researchers used the AI-designed robot as a blueprint. They first 3D printed a mold of the negative space around the robot’s body. Then, they filled the mold with liquid silicone rubber, letting it cure for a few hours. Once they popped it out of the mold, the silicone was sufficiently flexible. It was now time to test the robot’s walking skills.

First, they filled the rubber robot body with air, making its three legs expand. As the air was released from the robot’s body, the legs contracted. Through a continuous process of inflating and deflating the robot with air, it underwent a repetitive cycle of expansion and contraction, resulting in a gradual yet consistent form of locomotion.

“When humans design robots, we tend to design them to look like familiar objects,” Kriegman said in a news release. “But AI can create new possibilities and new paths forward that humans have never even considered. It could help us think and dream differently. And this might help us solve some of the most difficult problems we face.”

While the robot can’t do much more than shuffle forward, the researchers believe there’s a world of possibilities for tools created by the same program. One day, similar robots might be able to walk around the rubble of a collapsed building, identifying trapped people, or they might be able to traverse sewer systems to diagnose problems.

The study was published in the journal PNAS.