MIT: Imaging system lets robots look into locked boxes

Researchers at the Massachusetts Institute of Technology (MIT) have developed an image recognition technology that enables robots to look through walls and into closed boxes. To achieve this, the scientists use signal waves in the millimeter range (mmWave), similar to Wi-Fi. Hidden objects are reconstructed and visualized from the reflection data obtained.

MIT calls the system “mmNorm”, as the scientists write in the paper “Non-Line-of-Sight 3D Object Reconstruction via mmWave Surface Normal Estimation” (PDF). The millimeter waves can penetrate various materials. These include plastic, cardboard, and walls made of plaster and stone. They then hit objects behind them, which reflect the signals. The signal reflections are picked up by the system and evaluated by an algorithm that reconstructs the shape of the objects.

These are estimates of the surface normals. The system not only estimates the position of a reflection in space, but also the direction from which the surface reflects a signal. This allows a more precise reconstruction of the object.

High reconstruction accuracy

As a result, the system achieves an astonishingly high reconstruction accuracy of 96 percent. At the same time, mmNorm can also visualize smaller objects with complex structures and curved shapes. By comparison, similar current systems have an accuracy of just 78 percent. They are often unable to reconstruct smaller objects accurately enough.

The MIT researchers have already found an application for mmNorm. The system could be attached to humanoid robots that walk around production halls and inspect products in (sealed) boxes, for example, and check for quality defects. The scientists have already tested that this works with a robotic arm to which mmNorm was attached. The system was moved around a hidden object, which could be reconstructed and visualized. Defects in the object could be detected in the virtual image.

(olb)