- Mmnorm reconstructs complex hidden shapes using Wi-Fi frequencies without touching the object
- Robots can now see indoors crowded using reflected signals of the surrounding antennas
- The MIT technique beat the precision of the current radar of 18% on more than 60 objects tested
In the environments where visibility is obstructed, such as boxes inside, behind the walls or under other objects, artificial intelligence could soon have a new way of getting ahead.
MIT researchers have developed a technique called MMNORM, which uses millimeter wave signals, the same frequency range as Wi-Fi, to rebuild hidden 3D objects with surprising precision.
“We have been interested in this problem for a long time, but we have struck a wall because the methods passed, when they were mathematically elegant, did not have where we had to go,” said Fadel Adib, principal of the study and director of the Kinetics Group signal.
Overcome radar limitations
The previous techniques are based on the rear projection, which produces low -resolution images and fail when applied to small occlus objects such as tools or utensils.
Researchers have discovered that the defect lies in monitoring a physical property known as specularity – how millimeter wave reflections behave like mirror images.
Instead of simply measuring where the signals bounce, Mmnorm estimates the direction of the surface, which researchers call the normal surface.
“Based on specularity, our idea is to try to estimate not only the location of a reflection in the environment, but also the direction of the surface at this stage,” said Laura Dodds, principal author on the newspaper.
By combining numerous estimates of this type from different antenna positions, the system reconstructs the 3D curvature of an object, distinguishing between the forms as nuanced as the handle of a cup or the difference between a knife and a spoon in a box.
Each antenna collects reflections with a variable force depending on the orientation of the hidden object.
“Some antennas could have a very strong vote, some may have a very low vote, and we can combine all the votes together to produce a surface normal which is agreed by all the antenna locations,” added Dodds.
This new approach reached a 96% reconstruction accuracy on more than 60 objects, surpassing existing methods which have only reached 78%.
The system worked well on wooden, plastic, glass and rubber objects, although it always fights with dense metal or thick barriers.
While researchers are working to improve material resolution and sensitivity, potential use cases increase.
In security scanning or military contexts, MMNORM could reconstruct the shape of the concealed articles without bags or opening boxes.
This capacity could be essential for robots supplied by AI in warehouse automation, research and rescue or even assisted life environments.
Via techxplore
