Bio

Laura Dodds is a PhD student at MIT advised by Prof. Fadel Adib. Her work focuses on designing novel wireless sensing technologies for enabling non-line-of-sight perception. She has received multiple awards including the MIT presidential fellowship, Amazon Robotics Thriving Stars Fellowship, best masters thesis in AI at MIT, best paper award at IEEE RFID, and best paper finalist at ACM MobiSys and IEEE RFID.

Non-Line-of-Sight 3D Object Reconstruction via Millimeter-Wave Surface Normal Estimation

Recent decades have witnessed significant advancements in machine perception with computer vision, enabling impressive capabilities in robotics, automation, augmented reality, and more. However, perceiving the world using cameras (or LiDARs) limits the vast majority of today’s systems to line-of-sight. As a result, they struggle in real-world environments with significant occlusions. For example, piece-picking robots cannot detect items buried under a pile, automated systems cannot inspect contents inside sealed packages, and augmented reality devices fail to account for occluded objects.

In my research, I investigate a fundamentally different approach for perception that extends to non-line-of-sight (NLOS) scenarios. I use emerging millimeter-wave (mmWave) technology, which is being deployed in 5G, 6G, and next generation WiFi, and in radars for self-driving vehicles. Unlike visible light, mmWave wireless signals can traverse everyday occlusions, including walls, cardboard, fabric, and plastic, making them well-suited for NLOS perception.

While mmWave signals have been used for coarse NLOS sensing, enabling high-fidelity 3D reconstruction of hidden objects has remained elusive. My research departs from decades of research in mmWave sensing to enable a fundamentally new approach for high-accuracy NLOS reconstruction, and demonstrates an impressive empirical leap in reconstruction accuracy. Specifically, my research directly estimates the surface of the hidden object by using mmWave reflections to estimate surface normal vectors.

Looking forward, this new capability of high-accuracy NLOS 3D reconstruction could unlock new capabilities spanning robust robotic perception in cluttered spaces, automated through-package inspection for shipping and logistics, and enhanced augmented reality systems with occlusion-aware environmental understanding.