Researchers use diffusion modeIs to design jumping robots

I spoke to CSAIL head Daniela Rus about generative AI a few years back. Among other things, we discussed the ways in which the buzzy technology was poised to impact the future of robotics. There were the usual bits, like how LLMs have caused many roboticists to rethink human-machine collaborations using natural language. For Rus, however, the potential for new robot designs was the truly interesting frontier.

We were speaking a bit in abstractions at the time – as people tend to do when discussing such concepts. But the lab’s recently published paper, ‘Generative-AI-Driven Jumping Robot Design Using Diffusion Models,’ offers more concrete insight into ways the technology can help create more efficient systems.

Here the technology was tasked with developing a robot that can jump higher and land safer. The final product, constructed through leveraging a combination of diffusion models and simulation, was able to spring up around two feet in the air, marking a 41% improvement over a solely human-designed version.

“We wanted to make our machine jump higher, so we figured we could just make the links connecting its parts as thin as possible to make them light,” says the paper’s co-author, Byungchul Kim. “However, such a thin structure can easily break if we just use 3D printed material. Our diffusion model came up with a better idea by suggesting a unique shape that allowed the robot to store more energy before it jumped, without making the links too thin. This creativity helped us learn about the machine’s underlying physics.”

The process was fed 500 possible designs, which were whittled down to a dozen. From there, simulation was utilized to determine efficacy. The AI then converted the design into a system that could be brought to life via 3D printing. Next up, it helped design a foot that decreased post-landing falling by 84%.

Co-lead author, Tsun-Hsuan Wang, says diffusion aided designs are the first step toward incorporating more generative AI into robotics. “We want to branch out to more flexible goals,” he notes. “Imagine using natural language to guide a diffusion model to draft a robot that can pick up a mug or operate an electric drill.”