But the biggest improvements weren’t on display at the suit’s unveiling last fall. Instead, they’re hidden away in the xEMU’s portable life-support system, the astro backpack that turns the space suit from a bulky piece of fabric into a personal spacecraft. It handles the space suit’s power, communications, oxygen supply, and temperature regulation so that astronauts can focus on important tasks like building launch pads out of pee concrete . And for the first time ever, some of the components in an astronaut life-support system will be designed by artificial intelligence.
Jesse Craft is a senior design engineer at Jacobs, a major engineering firm based in Dallas that was tapped by NASA to revamp the xEMU life-support system. For Craft and the hundreds of other engineers working on the project, this requires a careful balancing act between competing priorities. The life-support system has to be safe, obviously, but it also has to be light enough to fit the weight limits for the lunar lander and strong enough to withstand the intense g-forces and vibrations it will experience during a rocket launch. “It’s a really big engineering challenge,” says Craft.
Squeezing more stuff into less space with reduced mass is the kind of complex optimization problem that aerospace engineers deal with all the time. But NASA wants boots on the moon by 2024, and meeting that aggressive timeline meant that Craft and his colleagues couldn’t spend weeks debating the ideal shape of each widget. Instead, they’re piloting a new AI-fueled design software that can rapidly come up with new component designs.“We consider AI to be a technology that can do something faster and better than a trained human can do,” says Jesse Coors-Blankenship, the vice president of technology at PTC, the American company that made the software. “Some of the software technologies are things engineers are already familiar with, like structural simulation and optimization. But with AI, we can do it faster.” This approach to engineering is known as generative design. The basic idea is to feed the software a set of requirements for a component’s maximum size, the weight it has to bear, or the temperatures it will be exposed to and let the algorithms figure out the rest.
PTC’s software combines several different approaches to AI, like generative adversarial networks and genetic algorithms. A generative adversarial network is a game-like approach in which two machine-learning algorithms face off against one another in a competition to design the most optimized component. It’s the same technique used to generate photos of people who don’t exist . Genetic algorithms, by contrast, are analogous to natural selection. They generate multiple designs, combine them, and then take the best designs of the new generation and repeat. In the past, NASA has used genetic algorithms to design optimal—and bizarre—antennas.
“The machine’s iterative process is 100 or 1,000 times more than we could do on our own, and it comes up with a solution that is ideally optimized within our constraints,” says Craft. It’s especially helpful given that the final design of the space suit life-support system is still in flux. Even a small change to the requirements in the future could result in weeks of wasted work by engineers.