MIT Robots Gain «X-Ray Vision» Using Wi-Fi-Like Signals

Researchers at MIT have developed a technology that lets robots pinpoint the shape of items hidden inside packaging or behind partitions. The mmNorm system leverages millimeter-wave radio signals (mmWave) – the same kind used in Wi-Fi and 5G – to build detailed 3D models of obscured objects.

mmNorm works by emitting mmWave signals that penetrate materials like plastic, cardboard, and other non-metals. These waves bounce off hidden objects within a box or behind an obstacle. What sets mmNorm apart is its analysis: it doesn't just look at the intensity of the reflected signal, but crucially, also analyzes the angle (orientation) of the surface it bounced off. The system calculates "surface normals" – vectors perpendicular to the object's surface at every point. By combining normal data from countless points in space using specialized algorithms, mmNorm constructs a highly accurate model of the object's curvature and shape.

In tests involving over 60 complex items (forks, mugs, tools), mmNorm achieved 96% accuracy in reconstructing shapes. That’s a significant 18% jump over the best existing alternatives (78%). The system can even distinguish and accurately model individual objects packed together in a single box, like a spoon, fork, and knife. It effectively scans items made of wood, plastic, rubber, glass, and metal – provided the object itself isn't hidden behind metal.

The MIT team plans to boost the system's resolution, improve its handling of less reflective materials, and prepare commercial prototypes for real-world tasks in factories and warehouses. mmNorm tackles a long-standing robotics challenge: reliably detecting and identifying objects outside direct line of sight. Its high accuracy and use of standard signals pave the way for practical applications in automation.