MAPLE-RF: Localization with Partial Maps

Thu, 17 Sep 2026 00:00:00 +0000

I lead MAPLE-RF, an efficient probabilistic RF source-localization method for environments that are still being mapped. The core problem is common in robotic search: a receiver knows its own pose, but most of the surrounding floor plan may remain unobserved, including obstacles that could block or reflect the signal.

Your browser does not support embedded MP4 video.

The simulation follows a receiver along a recorded route while the observed map grows and the transmitter posterior is updated.

Inference before the map is complete. RF paths are aligned with map knownness, occupancy, visibility, receiver pose, bearing, and range before one network pass scores candidate source locations.

Belief accumulation during exploration. Current measurements and accumulated posteriors sharpen the source belief as the receiver moves and the observed map expands.

Method

MAPLE-RF aligns estimated path angles and SNRs with spatial channels for map knownness, occupancy, line-of-sight visibility, receiver pose, bearing, and range. A residual U-Net then scores every candidate transmitter location in one pass. Unlike a full-grid digital-twin query, inference does not ray-trace every candidate whenever the map or receiver pose changes.

Findings

Training across mixed levels of map coverage is essential for localization with unexplored space.
MAPLE-RF retained about 94% of its complete-map 1-m recall when 75-85% of the map was unobserved.
Fresh-query inference was about 218 times faster than full-grid general-purpose ray tracing in the evaluated configuration.
Bayesian accumulation along exploration routes placed more posterior probability near the source than the compared baselines.

The current study is simulation-based. Its role in the broader research program is to make posterior RF localization practical when a robot must localize and map concurrently, before a complete digital twin is available.

Paper: — first and corresponding author; IEEE ICRA 2027, under review.

Partial Maps | Haozhe Lei

MAPLE-RF: Localization with Partial Maps

Method

Findings