Intel’s Latest Sensor Can Learn New Tricks

At a BMX competition, a rider breezes through a series of connected tricks that spin past the crowd in the blink of an eye. Rather than have judges, audience, and athletes wait for slow-motion video replays, the stadium broadcasts the name of each trick and how well the rider performed it almost instantaneously, precisely recreating the motion of his bike in an animation to boot.

It sounds like the future, but that's basically what attendees watched last month at the Intel Developer Forum, an annual conference for new gadgets like the Curie module, a dime-sized sensor that Intel believes could change the world of action sports.

Onstage was a BMX bike mounted with two Curie modules—one below the seat, one on the handlebars. Each Curie module houses a 32-bit Intel Quark microcontroller and uses a six-axis sensor with a gyroscope and accelerometer to measure motion. Each was also wirelessly connected to a laptop. As the rider went through his demonstration, the sensors broadcast the bike's performance metrics to an overhead display for the audience. They also broadcast the name of the trick in real time, and that, according to Intel, is what sets the Curie module apart from other devices on the market.

Thanks to an internal neural network that recognizes data sequences, the Curie module can classify and relay the name of a trick in a few tenths of a second—significantly less time than it would take, say, a competition announcer do the same. To that, the Curie module could see applications at future action sports events to potentially benefit the announcers, the audience, and the judges.

"Intel is very good at taking things that are sometimes done in software and putting them in hardware," said Jerry Bautista, senior vice president and general manager of Intel's New Devices Group. The Curie module is the result of Intel wanting to create a device with an internal processor that could learn through repetition, the way a child learns to recognize the scent of apple pie—after making the dessert a few times, he associates the smell of cinnamon and apples baking with something good to eat. "The neural network is in the hardware. Because it's purpose-built silicon hardware, it's very power efficient and very high performing. We've put the modules on snowboards, and we're working with some of the half-pipe guys, and we've already demonstrated this on skateboards."

The module has two modes, learning and classification. In learning mode, the more repetitions of data sequences that the network receives—the more a rider sends it through a 540 tailwhip—the better its "understanding" and "knowledge" of the movement. Once a single module learns a trick, usually after three to five repetitions, the data parameters are infinitely extractable, meaning that any number of modules can download the data and instantly be able to recognize the trick. On classification mode, the module puts its education to use.

Intel had hired five riders to educate the modules in July and August, and the sensors on the BMX bikes at the conference recognized 14 ramp tricks and 10 tricks on flat ground. Wesley Monroe was one of those riders and says that the real-time flow of data helped him better understand how small adjustments to his riding could change his performance.

"One thing it did was help you know how you could change things to make it easier on your body," he said. "If you landed a jump pretty hard, you could see that maybe coming in a little faster would give you a softer landing."

Bautista and others think the Curie has wide applicability in sports medicine, specifically in measuring wear and tear on athletes. For instance, the sensor can measure the stress that a BMX rider absorbs through his feet, ankles, and legs, which can mean g-forces in the high teens and low 20s, according to Bautista. Skateboarders, Intel found, can sustain impacts of more than 30 g's when landing a trick.

Intel will begin sending the modules to manufacturers sometime this year; however, the devices won't be available commercially until the FCC approves it. Intel, though, is not in a hurry to implement the Curie module on a large scale.

"The other part of this that's interesting is the notion of unintended consequences," Bautista said. "A lot of times the technology can run ahead of the social and security aspects of things."

Intel recently put Curie modules in the helmets of college football players to measure impact forces during tackling, which can lead to concussions. Bautista and others were trying to determine what type of tackles effectively stopped the runner without injuring either player, and the sensors recorded data that could show concussive and "pre-concussive" behavior.

"This sounds like a great thing, right?" Bautista said. "The coaches liked the data, the sports medicine staff liked the data, but the players did not like it at all. And the simple reason was that it was starting to show when they were getting concussions, and they didn't want the medical staff or the coaches to know that."

The protocol for a player with a concussion is to sit out the following game, and the players didn't want to do that, Bautista said. From their perspective, the new data put them at risk of losing a starting lineup position or missing a game that an NFL scout might be watching.

"So these guys wanted to hide it, and it brought up this whole notion of who owns the data," Bautista said.

The school claimed it owned the data because it owned the equipment issued to players, and the helmet was just another piece of equipment, with or without a sensor in it. The players claimed they owned the data because it was their bodies taking the beating. Intel then discontinued using the module in football helmets.

"We're trying to go about this in a very deliberate way," Bautista said. "We want to introduce [the Curie module] in such a way that doesn't disrupt or damage these sports, but rather enhance them"

With its dime-sized module, Intel is fast approaching the wearable technology market. Last week it debuted a sensor-equipped sports bra that increased ventilation when body temperature increased. Actions sports is just a single facet of the technology's potential, and one that the company continues to devote time and money to.

"The fundamental assumption here is that if you don't measure it, you can't improve it," Bautista said. "If we're not measuring the kinds of things that are impacting athletes, we can't improve them."