Full Breakdown
Smartphone LiDAR Achieves Non-Line-of-Sight Imaging
5/22/2026, 1:00:28 PM
MIT Media Lab Demonstrates NLOS Imaging with Consumer LiDAR
Researchers led by Siddharth Somasundaram at MIT Media Lab have shown that a smartphone-grade LiDAR sensor can reconstruct hidden 3-D objects, track motion around corners, and infer camera pose. The system uses a ~100-pixel LiDAR module and a motion-induced aperture sampling algorithm that merges many noisy frames captured while the device moves.
Background: From Expensive Labs to <$100 Hardware
Non-line-of-sight (NLOS) imaging has traditionally required ultrafast lasers, high-sensitivity detectors, and careful calibration, often costing tens of thousands of dollars. Prior work achieved high-resolution reconstructions in specialized labs. The MIT approach replaces these components with off-the-shelf LiDAR hardware under $100, eliminating precise alignment.
Key Contributors
Siddharth Somasundaram – Lead researcher, MIT Media Lab; Jessica Rosenworcel – Executive director, MIT Media Lab.
Technical Highlights
Sensor: ~100-pixel LiDAR with picosecond time-of-flight.
Algorithm: motion-induced aperture sampling fuses dozens of frames, akin to burst photography and synthetic aperture radar, to amplify weak multi-bounce signals.
Demonstrated: reconstruction of static hidden objects (e.g., a U-shaped form), tracking of multiple moving objects, and camera pose estimation from hidden landmarks.
Limitations: sparse “probability clouds” rather than crisp images; performance drops on diffuse surfaces, favoring reflective targets.
Why It Matters
The ability to sense around corners with cheap LiDAR could reshape robotics (anticipating obstacles), AR/VR (maintaining hand tracking beyond view), autonomous driving (early detection of hidden pedestrians), and broader machine vision, as developers explore the released code and data.
Official Statements & Responses
Somasundaram said the work democratizes NLOS imaging, making it accessible to robotics, AR/VR, and consumer devices. Rosenworcel called the achievement the realization of a long-standing vision, noting that perception will expand in unforeseen ways.
Criticism & Limitations
The system’s low spatial resolution and reliance on reflective surfaces limit use. Extending indoor demos to outdoor settings, especially for autonomous vehicles, will require handling brighter ambient light, motion blur, and safety certification.
Conflicting Reports & Gaps
All sources agree on the sub-$100 hardware and novel algorithm, but differ on performance: some highlight successful multi-object tracking, others note reduced fidelity on diffuse surfaces. No quantitative error metrics are provided.
Verbatim Quotes
- “The most exciting part of this work to me is that we took a capability that used to require a specialized $50,000 imaging setup and put it into the hands of people in robotics, AR/VR, and beyond,” — Siddharth Somasundaram, MIT Media Lab
- “Once we developed algorithms that could combine information across those measurements, the hidden signals started to emerge much more clearly,” — Siddharth Somasundaram, IEEE Spectrum
- “We think the most important implication is the democratization of the technology,” — Siddharth Somasundaram, IEEE Spectrum
- “Now that vision is arriving in consumer devices, with implications we’re only beginning to explore,” — Jessica Rosenworcel, MIT Media Lab
What’s Next
The team released code and data, inviting the community to improve robustness on diffuse surfaces and integrate the approach into AR headsets and autonomous platforms.
