SuperEx: Enhancing Indoor Mapping and Exploration using Non-Line-of-Sight Perception

TL;DR

SuperEx tackles indoor mapping and exploration under occlusions by integrating non-line-of-sight cues from single-photon LiDAR into a three-stage pipeline: physics-based SPAD histogram simulation, Pix2Pix-based back-projection filtering, and LaMa-powered global map prediction fused with probabilistic frontier planning. The approach enables robots to see around corners, carving empty NLOS regions and reconstructing occupied structures to guide exploration. On real-world KTH Floorplan data and simulated maps, SuperEx achieves higher coverage and IoU than LOS baselines, with substantial early-stage gains when visibility is limited. This work demonstrates the practical feasibility of NLOS perception in autonomous exploration, paving the way for more reliable mapping in cluttered indoor environments.

Abstract

Efficient exploration and mapping in unknown indoor environments is a fundamental challenge, with high stakes in time-critical settings. In current systems, robot perception remains confined to line-of-sight; occluded regions remain unknown until physically traversed, leading to inefficient exploration when layouts deviate from prior assumptions. In this work, we bring non-line-of-sight (NLOS) sensing to robotic exploration. We leverage single-photon LiDARs, which capture time-of-flight histograms that encode the presence of hidden objects - allowing robots to look around blind corners. Recent single-photon LiDARs have become practical and portable, enabling deployment beyond controlled lab settings. Prior NLOS works target 3D reconstruction in static, lab-based scenarios, and initial efforts toward NLOS-aided navigation consider simplified geometries. We introduce SuperEx, a framework that integrates NLOS sensing directly into the mapping-exploration loop. SuperEx augments global map prediction with beyond-line-of-sight cues by (i) carving empty NLOS regions from timing histograms and (ii) reconstructing occupied structure via a two-step physics-based and data-driven approach that leverages structural regularities. Evaluations on complex simulated maps and the real-world KTH Floorplan dataset show a 12% gain in mapping accuracy under < 30% coverage and improved exploration efficiency compared to line-of-sight baselines, opening a path to reliable mapping beyond direct visibility.

Figures (9)

• Figure 1: Non-line-of-sight (NLOS) perception enables efficient exploration. Measuring and understanding NLOS light paths enables the robot to distinguish a dead-end from an open corridor, a challenging task to perform from line-of-sight vision alone; and explore the unknown indoor environment efficiently.
• Figure 2: Principle of NLOS sensing: Single-photon LiDAR comprises a pulsed laser, single-photon detector, and timing circuits. (a) When the laser pulse strikes a visible wall, it diffuses, and some of the scattered rays hit the hidden object. Some of the light is scattered back and captured by the sensor as time-of-flight histograms (b), recording the number of photons in each time bin. These measurements are then converted into back-projection maps (c), which represent the likelihood of an object’s presence at a certain distance from the wall.
• Figure 3: SuperEx pipeline. The histograms captured by the single-photon LiDAR enable 1) carving out NLOS regions that are empty and 2) backprojection of occupied NLOS, that is filtered with a Pix2Pix network. Both the carved occupancy and filtered backprojection are fed into the Lama network for improved global map prediction, and then for enhanced frontier exploration.
• Figure 4: Back projection map filtering and global map prediction: Using the filtered back projection map and the observation map, pipeline predicts a plausible global map.
• Figure 5: Filtered back-projection map during NLOS sensing improves global map prediction over LOS sensing. (a) As the robot is traversing, an additional obstruction is introduced. (b) With LOS sensing, global map prediction fails to reconstruct the additional obstacle, resulting in an open corridor for exploration. (c) With NLOS sensing, we obtain a filtered back-projection map that correctly reconstructs the additional obstacle. (d) Global map prediction for NLOS takes this filtered back-projection map and correctly predicts the dead end.