X-ray vision is one of the world’s most popular superpowers. Young boys all over the world, at some point, even wished to have this superpower but could only resign to daydreaming about it.
Presently, that is about to change because, for the past ten years, scientists from the Computer Science and Artificial Intelligence Laboratory at MIT are doing their best to bring us closer to seeing through walls.
AI that Equates to Having a Superpower
The “RF-Pose”, which is the latest project of a team of scientists at MIT, uses artificial intelligence in order to instruct wireless devices to sense people’s poses and gestures, even when they are obstructed from view.
The team uses a neural network in order to analyze radio signals, which usually bounce off people’s bodies, and then generate a dynamic stick figure that walks, sits, and stands as the person performs those actions. Moreover, the system doesn’t even have to see to know how someone is walking, sitting, or standing.
In order to train the neural network, the MIT scientists collected samples of people walking using both wireless signal pings and cameras. Now, those camera footages were processed to create stick figures to substitute real people.
The scientists complemented that data with radio waves, and that combined data was used by researchers to teach the neural network. With a deep-seated connection between the stick figures and the RF data, stick figures are successfully created based on radio wave reflections.
Fascinatingly, the camera can’t actually see through the walls. So, it is clear that the system was never explicitly trained in recognizing people on the other side of an obstacle. The only reason that it works is that the radio waves bounce off a person on the other side of a wall, just like how they would do when they’re in the same room. Moreover, the system also works with many people crossing paths.
Overcoming the Challenges
One of the hurdles that the scientists had to overcome is that a majority of neural networks are trained using data labeled by hand; for instance, a neural network that is taught to recognize dogs needs people to look at a big dataset of images and label each one as either “dog” or “not dog”. Manually labeling radio signals, on the other hand, proves to be quite difficult.
So as to deal with this concern, the scientists gathered images of people doing activities like standing, walking, sitting, and opening doors. Afterward, they used these images to generate stick figures, which they presented to the neural network together with the matching radio signal. This enabled the system to learn the connection between the radio signal and the stick figures of the people in the scene.
Practical Uses of RF-Pose
According to the scientists, there are quite a few practical applications for RF-Pose; one of them is to monitor diseases, such as Parkinson’s and muscular dystrophy, in order to provide a better understanding of the disease and will allow doctors to make modifications to the medications correspondingly.
The scientists also said that RF-Pose could also help elderly people live more independently by offering an extra security of monitoring for falls, injuries, and changes in activity patterns.
At present, the team is working with doctors to discover future applications of RF-Pose in health care. Apart from health care, the scientists state that RF-Pose could possibly be used for new classes of video games and also in search-and-rescue missions that would need help in locating survivors. Now that’s what I would call a superhero power.