Active Sensing for Robust Perception in Adverse Conditions

20 concepts20 papers1 proposal

Most robotic perception systems rely on passive vision—cameras that capture ambient light reflected off surfaces. While effective in well-lit, textured environments, this dependency makes robots vulnerable to the "real world": darkness, fog, glare, and featureless corridors. A growing body of research suggests that active sensing, primarily through LiDAR and laser range finders, provides a fundamental solution to these reliability gaps by decoupling perception from environmental lighting.

The Advantage of Active Illumination

The core intuition behind active sensing is control. Unlike a camera, which is a passive observer of available photons, a LiDAR sensor emits its own signal and measures the return. This makes the system inherently invariant to external lighting conditions.

As Triess et al. note, this active nature renders the "day-to-night" domain adaptation problem—a major hurdle for visual systems—almost trivial for LiDAR-based models. Because the sensor provides its own light source, it maintains consistent performance across different times of day without needing specialized algorithms to handle shadows or low light. This stability is echoed by Paek et al., who demonstrated that LiDAR-based lane detection shows negligible performance degradation between day and night, whereas visual sensors frequently fail due to glare or total darkness.

This reliability extends to atmospheric occlusions. Kim et al. found that laser-based systems remain "rock-solid" in thick fog where cameras become blinded, simply because the active signal can penetrate or be distinguished from environmental noise more effectively than ambient light.

Stability in Feature-Poor Environments

Beyond lighting, robotic systems often struggle with "texture-poor" environments—places like underground garages, sterile hospital hallways, or dusty construction sites where visual landmarks are repetitive or non-existent.

In these scenarios, the geometric precision of active sensing outperforms visual patterns. Hou et al. observed that in underground parking structures, visual SLAM (Simultaneous Localization and Mapping) suffers from high trajectory errors due to repetitive textures and dim light. LiDAR-inertial approaches, however, maintain high robustness by mapping the physical structure of the space rather than its visual appearance. Similarly, Helmberger et al. highlight that in industrial settings, LiDAR-inertial fusion provides significantly higher accuracy because it does not rely on the distinct visual landmarks that construction sites often lack.

Overcoming Motion and Appearance Changes

Active sensing also mitigates challenges related to the robot's own movement and the appearance of the subjects it tracks. Visual sensors are prone to motion blur during rapid maneuvers, a problem Wu et al. addressed by using LiDAR-centric perception for high-speed quadrupedal robots. By using distance measurements rather than light-sensitive pixels, the robots could maintain localization during aggressive soccer maneuvers that would blur a standard camera frame.

Furthermore, active sensors are less sensitive to changes in the appearance of objects. Chalvatzaki et al. found that laser-based sensors were superior for tracking human gait because they are unaffected by variations in clothing or ambient lighting that might confuse a visual person-tracker. This concept is pushed further by Guo et al., who demonstrated that LiDAR-based person re-identification can achieve high accuracy even in total darkness by focusing on 3D point cloud geometry rather than 2D visual features.

Intensity as a Lighting-Invariant Image

While LiDAR is often prized for its 3D structural data, the intensity of the laser return offers a unique middle ground between vision and geometry. The authors of Robust Place Recognition using an Imaging Lidar describe a method of projecting LiDAR intensity into 2D "images." Because this intensity is based on the reflectivity of the material to the laser rather than ambient light, the resulting "images" are invariant to external illumination.

This allows robots to perform loop closure and place recognition in scenarios ranging from direct sunlight to pitch blackness using the same features. Mutz confirms this advantage in the context of self-driving cars, finding that reflectivity-based maps are a "gold standard" for precision because they remain stable across all environmental conditions, unlike color-based maps which are too unstable for reliable long-term use.

Key Foundational Work

A Real Time Lane Detection Algorithm Using LRF for Autonomous Navigation of a Mobile Robot (2013) — contributed 1 concept elements to this direction.
Towards an intelligent robotic walker for assisted living using multimodal sensorial data (2014) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_sensor_modality_robustness_cf5537b7,
  title = {sensor modality robustness},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/sensor-modality-robustness/rdr_cf5537b769fa45d29441461cd73de4a7},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_sensor_modality_robustness_cf5537b7_v1,
  title = {sensor modality robustness},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/sensor-modality-robustness/rdr_cf5537b769fa45d29441461cd73de4a7/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake