One forward-facing radar maps furniture on both sides of a home robot's path

Researchers led by Yingqi Wang have described a way for a mobile robot to build a 3D picture of furniture and structures on both sides of its path with a single forward-mounted millimetre-wave radar, in a paper posted on arXiv.

Close-range wavefront curvature, targets that are visible only from some viewpoints and limited spread in height make this hard, the authors write. Their method uses the robot's own motion: a near-field model links the radar's stepped-frequency MIMO measurements across frequencies and robot positions, sliding-aperture back-projection focuses them, and images with different visibility are combined before heights are extracted and refined.

In controlled, office or study, and corridor settings, the mean lateral-forward localisation errors were 4.65, 5.63 and 5.98 cm, with peak-to-background ratios above 13 dB. The method needs one forward pass and does not rely on learned models of the scene.