According to Adam Tilton, when you get down to it, there really isn’t that much of a difference between estimating the path and speed of a missile and figuring out what kind of exercise you’re doing at the gym: it’s all about using a sensor to measure a signal, and extracting that signal from the surrounding noise.
Tilton should have a pretty good idea of what he’s talking about. As a graduate student at the University of Illinois at Champaign-Urbana, he studied signal processing and estimation control—typically applied to things like guiding missiles. But a few years ago, after realizing the technology could also be useful for the fast-growing market for wearable gadgets like fitness-tracking wristbands and smart watches, he decided to change course.
Last year, Tilton and his PhD advisor Prashant Mehta, an associate professor of mechanical science and engineering, started working on Rithmio, a software startup that seeks to make it easier to add accurate gesture and activity recognition to all kinds of wearable gadgets that can quickly learn new activities without needing lots of training data.
The market for wearables has grown rapidly in recent years; according to data from technology market researcher IDC, shipments of wearable devices tripled in the first three months of this year to 11.4 million, compared with 3.8 million a year earlier. The most popular wearables are for fitness tracking, such as those from Fitbit, Xiaomi, and Garmin. But gadgets currently on the market tend to track just a handful of gestures and activities like running, stair climbing, cycling, and sleep, and need to be trained with lots of data in order to learn how to identify activities or gestures.
Rithmio’s founders think they can make this better, faster, and easier—which could encourage more people to try wearables and, over time, keep using them for everything from exercise tracking to using gestures to control other gadgets like TVs and tablets. Tilton, Rithmio’s CEO, says its software can learn all kinds of new, specific motions: bench presses, jumping jacks, biceps curls, and so on—in seconds, right on a user’s wrist.
“That alleviates the training problem,” he says.
The Chicago-based startup also recently raised $3 million in venture-capital funding from chip maker Intel’s venture investment arm, Intel Capital, KGC Capital, and others.
So far, Rithmio has built (but not yet released) several apps that can do things like track weight-lifting activities, and it plans to release software tools later this year so developers can add its gesture-recognition capabilities to smart watches that run Google’s Android Wear operating system and to the Apple Watch and iPhone.
And Tilton says the company is also working with chip makers such as Intel and smart watch and smartphone makers to build its software into chips and devices. He hopes Rithmio will be incorporated into products sometime next year, though he doesn’t say what form it may initially take.
The company has been testing its technology on a variety of Android Wear smart watches like the Sony SmartWatch 3 and Motorola 360, and it has tracked more than 30 different gestures so far, he says. It can learn new ones in six to 10 seconds.
Tilton won’t give many technical details, but he says that unlike typical gesture and activity recognition, Rithmio’s method doesn’t need to do things like analyze lots of movement data to find a set of features that make up an activity (like walking) or use a classifier, which is software that recognizes the activity when it happens.
A video demo featuring Tilton gives a sense of how Rithmio’s gesture learning works. In it, he wears a wrist-based motion sensor on his left arm that includes an accelerometer and a gyroscope, and he does several jumping jacks—an activity that the software (shown on a nearby laptop screen) doesn’t yet know. Each new activity is marked as a “new rhythm,” Tilton says, and the user can then give each one a proper name.
Tilton says that once Rithmio’s software learns a new activity—squats, bicep curls, or bench presses, for instance—it can also gauge metrics like speed and repetition. In addition, its algorithm can tell the difference between nearly identical actions, like small- versus medium-sized arm rotations. He says Rithmio is also working on determining things like your form when performing an exercise and efficiency as compared to past workouts.
While the market for wearable tech is still largely focused on the wrist, Tilton says, Rithmio will work on different parts of the body and in concert with other sensors on the body that also run the software (this could make Rithmio even more accurate).
Tilton says Rithmio has also built a demo app that physical therapists could use to track how well and often patients are doing assigned exercises.
Yet a key difficulty for Rithmio will be the same one facing all kinds of companies in the wearables market: simply getting people to wear gadgets on their bodies and track what they’re doing. Despite the fast growth of wearables, the number of people trying them out is still small, and it’s not clear how making it easier to add more activities to a gadget will convince more to join in.
These weird virtual creatures evolve their bodies to solve problems
They show how intelligence and body plans are closely linked—and could unlock AI for robots.
Surgeons have successfully tested a pig’s kidney in a human patient
The test, in a brain-dead patient, was very short but represents a milestone in the long quest to use animal organs in human transplants.
A horrifying new AI app swaps women into porn videos with a click
Deepfake researchers have long feared the day this would arrive.
The covid tech that is intimately tied to China’s surveillance state
Heat-sensing cameras and face recognition systems may help fight covid-19—but they also make us complicit in the high-tech oppression of Uyghurs.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.