Fantastic: Artificial intelligence designs unique walking robots in tens of seconds

Groundbreaking artificial intelligence (AI) has astoundingly completed the design of a functional walking robot in a matter of seconds, demonstrating rapid evolution in stark contrast to nature's multibillion-year journey.

This AI runs on an ordinary personal computer to build a completely innovative structure from the ground up, unlike other AI models that rely on massive amounts of data and high-performance computing. Emerging from a simple "design a walker" prompt, the robot evolved from an immovable block to a strange, porous, three-legged entity capable of slow, steady movement.

This AI-designed organism represents more than just a mechanical achievement, it could mark a demonstrative shift that provides a novel, unfettered perspective on design, innovation, and potential applications in fields ranging from search and rescue to medical nanotechnology.

Key findings:

1. Rapid Design: AI created a functional walking robot that evolved its design from a static block to a moving entity in just 26 seconds on a basic laptop, demonstrating the unprecedented speed of robot design and evolution.
2. Innovative structure: Unlike traditional AI, which tends to mimic existing designs, the system makes an entirely new, oddly shaped, and effective robot with three legs and a porous body, revealing a solution without human design bias.
3. Practical and future applications: The impact and applications of such AI-driven design are enormous, encompassing potential developments in various areas of the future, such as crisis-response robots capable of navigating through the rubble to locate survivors, or nanorobots designed to traverse the human body for medical diagnosis and treatment.

A team led by researchers at Northwestern University has developed the first artificial intelligence (AI) to date that can intelligently design robots from scratch.

To test the new AI, the researchers gave the system a simple hint: Design a robot that can walk on flat surfaces. While it took nature billions of years to evolve the first walking species, the new algorithm compresses evolution to lightning speed — designing a successful walking robot in just a few tens of seconds.

But AI programs aren't just fast. It can also run on a lightweight personal computer and design a completely new structure from scratch. This is in stark contrast to other AI systems, which typically require energy-hungry supercomputers and huge data sets. Even after processing all this data, these systems are still limited by human creativity – they can only mimic human past work, without the ability to generate new ideas.

The study will be published October 3 in the Proceedings of the National Academy of Sciences.

Sam Kriegman of Northwestern University, who led the work, said: "We found a very fast AI-driven design algorithm that can bypass evolutionary traffic jams without relying on the biases of human designers.

We told AI that we wanted a robot that could travel across land. Then we just have to press a button and that's it! In the blink of an eye, it generated a blueprint of a robot that looked very different from any animal that had ever walked on Earth. This process is called 'instant evolution'. ”

Krigman is an assistant professor of computer science, mechanical engineering, and chemical and biological engineering at Northwestern's McCormick School of Engineering, where he is a member of the college's Center for Robotics and Biosystems. He worked closely with other researchers Matthew Ann, as well as Drew Spielberg of MIT, Daniela Ross of the University of Vermont, Josh Bongard and others for many years to make these groundbreaking discoveries.

From xenobots to new creatures

In early 2020, Krigman attracted a lot of media attention for developing Xenobots, the first living robot made entirely from biological cells. Now, Krigman and his team see their new AI as the next step in exploring the potential of artificial life. The robot itself is unremarkable – small, soft, and strangely shaped. Currently, it is made of inorganic materials. But Krigman said it represents the first step in a new era of AI design tools that, like animals, act directly on the world.

When people see this robot, they may think that they are just seeing a useless gadget, but these scientists discover the birth of a completely new organism.

Go from zero to walk in seconds

While the AI program can be launched at any prompt, Krigman and his team started with a simple request to design a physical machine capable of walking on land. This is where the input from the researchers ends and is taken over by artificial intelligence.

The computer starts with a block the size of soap. It can shake, but definitely not walk. The AI knew it hadn't reached its goal, so it quickly iterated on the design. In each iteration, the AI evaluates its design, identifies defects and cuts simulation blocks to update its structure.

New AI algorithms compress billions of years of evolution into seconds (Credit: Neuroscience News)

Eventually, the simulated robot can bounce in place, then jump forward, and then drag its feet. Eventually, after nine attempts, it produced a robot that could walk half the length of its body per second — about half the average human stride speed.

The entire design process – from a zero-motion invisible block to a complete walking robot – took just 26 seconds on a laptop.

Now, as AI generates better and better robotic bodies in real time, anyone can watch the evolution.

Evolutionary robots previously required weeks of trial and error on supercomputers, and of course, billions of years of trial and error before any animal could run, swim, or fly in our world. This is because evolution has no vision. It cannot foresee the future and know whether a particular mutation will be beneficial or catastrophic. We've found a way to remove this blindness, thus compressing billions of years of evolution into moments.

Rediscover your legs

Surprisingly, the AI alone came up with the same walking solution as nature: legs. But unlike the absolutely symmetrical design of nature, AI takes a different approach. The resulting robot has three legs, fins on its back, and a flat face covered with holes.

Interestingly, because the researchers haven't told AI that the robot should have legs, it has been able to spot that legs are a good way to move on land. In fact, leg exercise is the most effective form of land exercise.

To see if the simulated robot could work in real life, Krigman and his team used an AI-designed robot as a blueprint. First, they 3D printed a mold of the negative space around the robot's body. They then fill the mold with liquid silicone rubber and let it cure for several hours. When the team removed the cured silicone from the mold, it became soft and flexible.

Now, it's time to see if the robot's simulated behavior (walking) remains in the physical world. The researchers filled the rubber robot's body with air, inflate its three legs. When air is expelled from the robot's body, the legs contract. By constantly pumping air into the robot, it expands and then contracts repeatedly, resulting in a slow but steady movement.

Unfamiliar design

While the evolution of the legs makes sense, these holes are a strange addition. The AI punched holes in seemingly random places on the robot's body. Krigman hypothesized that porosity could reduce weight and increase flexibility, allowing the robot to bend its legs to walk.

Researchers don't know much about what these holes do, but it's important to know them because when they remove them, the robot either can't walk anymore or can't even walk.

Overall, Krigman was surprised and fascinated by the robot's design, noting that most human-designed robots either look like humans, dogs, or hockeys.

When humans design robots, we tend to design them to look like familiar objects, but AI can create new possibilities and new paths forward that humans have never considered, Krigman said. It can help us think and dream differently. This may help us solve some of the most difficult problems we face. ”

Potential applications for the future

Although AI's first robot could only move forward, Krigman imagined a world of possibilities with tools designed by the same program.

One day, similar robots might be able to navigate the rubble of collapsed buildings, following heat and vibration signals to search for trapped people and animals, or they might traverse sewer systems to diagnose problems, unclog pipes and repair damage.

AI may also design nanobots that enter the human body and direct blood flow to unblock arteries, diagnose diseases or kill cancer cells.

The only barrier to people using these new tools and therapies is that we don't know how to design them, and now fortunately, AI has its solution.