Apparently Aladdin now has more choices. All kidding aside, while it can't speak or fly, this MIT-developed tactile sensing mat has something (scientifically) magical about it.
Il MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) He developed an intelligent carpet that can observe, evaluate and catalog every type of position assumed when you walk or sit on it. A step forward towards extremely interesting products in the field of smart home, remote healthcare and why not, video games.
A carpet that assists us
Many of our daily activities involve physical contact with the ground: walk, exercise, rest. These integrated interactions contain a wealth of information that helps us better understand people's movements.
Previous research has exploited the use of single RGB cameras (from the old Microsoft Kinect onwards), but not only. Wearable omnidirectional cameras, and even webcams (with all the inevitable privacy issues) have also been employed. What's different about this project?
To deduce the 3D pose with the CSAIL team system, a person simply has to step onto the mat and perform any action. The neural network developed by the team determines what someone is doing just by using tactile information. She's sitting? He is walking? Is she lying down?
Tomorrow he will tell us: are you doing yoga well or badly? Do you have a hip problem? Do you sleep properly?
Imagine taking advantage of this model to enable a seamless health monitoring system. To detect falls, for example, or to monitor and assist with motor rehabilitation.
Yiyue Luo, lead author of the study
The low-cost, scalable carpet was made with a pressure-sensitive film and conductive thread. It incorporates over nine thousand sensors that cover 10 square meters of carpet (a size larger than the average of any furniture carpet).
How does the tactile sensing mat work?
Each of the sensors on the carpet converts human pressure into an electrical signal, through physical contact between people's feet, limbs, torso and carpet. The system has been specifically trained on synchronized tactile and visual data, such as a video and the corresponding heat map of someone doing a pushup.
The model takes the pose extracted from the visual data, uses the tactile data as input, and finally shows the human pose in 3D.
Indeed, the carpet was able to predict the pose of a person with a margin of error of less than ten centimeters. For the classification of specific actions, then, even the system was accurate 97% of the time.
A coach mat
As mentioned, one of the possible applications is linked to the monitoring of physical activity. Imagine that the mat recognizes our activity, counts the number of repetitions we do (for example for push-ups) and also calculates the corresponding calories burned.
Current limits? The obvious ones: Since much of the pressure distributions on a mat are induced by lower body and torso movement, this information is currently more accurate than upper body data.
Future improvements? If a carpet can do this with just one person, imagine an entire surface covered in these sensors. An entire nightclub could detect the state of lucidity of all its guests by the coherence of their movements. Something like many others, the only limit is your imagination.