What are the trade-offs of using 3D Point Cloud vs Stereo Vision for autonomous robot navigation?

All answers to this question.

Reflective surfaces are the "kryptonite" of LiDAR. Most LiDAR sensors rely on Time-of-Flight (ToF), and polished concrete can cause beams to bounce away, creating "ghost" objects or holes in your map. Stereo Vision, while computationally more intensive, is generally better at handling reflections because it relies on visual feature matching. However, for a warehouse, a "Sensor Fusion" approach is the real industry standard for 2024. You use a 2D LiDAR for high-speed floor-level obstacle detection and a depth camera (Stereo) for 3D object recognition. This way, if the LiDAR gets confused by a chrome pillar, the visual system provides the necessary redundancy to prevent a collision.

   Answered 2025-07-10 by Barbara Richardson

Does the lighting requirement for Stereo Vision become an issue in dark warehouse corners, or do modern active-stereo cameras solve that with IR projectors?

   Answered 2025-07-25 by Daniel Evans

• Daniel, active stereo cameras like the Intel RealSense or OAK-D use infrared projectors specifically to solve the "textureless" or "dark" environment problem. They project a pattern onto the scene, which the cameras use to calculate depth even in total darkness. This makes them much more versatile than traditional passive stereo setups and a very strong competitor to low-resolution LiDAR sensors in an indoor AGV application.

     Commented 2025-08-05 by Joseph Miller

We found that LiDAR is still easier for SLAM (Simultaneous Localization and Mapping). The point clouds are much simpler to process for loop closure than visual features.

   Answered 2025-08-15 by Patricia Lewis

• I agree, Patricia. LiDAR SLAM is very mature. If you can handle the reflection issues with some clever filtering, it definitely saves on the processing power needed for the navigation stack.

     Commented 2025-08-20 by Kevin Parker