LiDAR vs. Stereo Cameras for Humanoid Navigation

Two very different ways to build a 3D map of a room, with different strengths for a walking robot.

LiDAR measures distance by timing how long laser pulses take to bounce back off surfaces, building a precise 3D point cloud largely independent of lighting conditions. Stereo cameras estimate depth by comparing the same scene from two slightly offset lenses, the way human binocular vision works, computing distance from the disparity between the two images.

Where each wins

LiDAR is generally more accurate at range and unaffected by darkness, but tends to be heavier, pricier, and lower-resolution for fine detail than a good stereo camera pair. Stereo cameras are cheap, compact, and high-resolution, but struggle in low light and on featureless surfaces (a blank white wall gives the matching algorithm nothing to compare).

Why many humanoids use both

A lot of platforms pair a lower-resolution LiDAR or depth sensor for robust obstacle detection with stereo or RGB-D cameras for richer scene understanding and object recognition, leaning on each sensor's strengths rather than picking one exclusively.