IndoorBEV: A Lightweight Real-Time LiDAR BEV Perception System for Indoor Mobile Robots

What happened

arXiv:2610.00355v1 Announce Type: new Abstract: Efficient indoor LiDAR (a sensor that measures distance with laser pulses) perception is challenging because mobile robots must understand cluttered three-dimensional environments under strict latency (the delay between asking for something and getting it) and memory constraints. A lightweight encoder then integrates complementary geometric features with multi-scale local representations and compact global scene context.

Existing point-based and voxel-based methods often incur substantial computational overhead, whereas conventional bird's-eye-view (BEV) representations improve efficiency at the cost of discarding vertical geometric information. IndoorBEV summarizes the vertical point distribution in each BEV cell using statistical height features and multi-frequency height encoding, allowing informative three-dimensional cues to be processed efficiently by two-dimensional convolutions.

Decoupled dense prediction heads jointly produce semantic BEV maps and oriented object bounding boxes. IndoorBEV contains only 0.6M parameters and requires 2.3 MB of model storage.

Key facts

• IndoorBEV — includes: only 0.6M parameters and requires 2.3 MB of model storage