A robot may have a floorplan, yet still need a huge prepared database before it can tell where it is. FreeLoc asks a simpler question: can the robot query the map directly, as it moves? The practical challenge is this: given only a floorplan, a robot should be able to query it online and achieve accurate camera-based localization in new environments, without constructing a scene-specific offline database.

That matters because the robot would not need a prepared map database for every new place before it could begin locating itself. FreeLoc treats the floorplan as a directly queryable geometric map rather than a scene-specific offline retrieval database. Think of it like asking a paper map for directions at the moment you need them, instead of photographing every possible route in advance and storing all those photographs.

FreeLoc first retrieves a compact set of plausible pose anchors by querying geometric rays from the floorplan on the fly. These anchors are rough possibilities for where the camera might be, rather than a dense list of every possible position. To refine these approximate anchors, FreeLoc estimates residual pose corrections from the discrepancy between image-side and floorplan-side ray observations.

Instead of relying on one-shot deterministic regression, pose refinement becomes an iterative denoising process that progressively corrects imperfect coarse anchors under geometric guidance. This lets FreeLoc move beyond the resolution of coarse online samples and produce accurate pose hypotheses in continuous pose space, without exhaustive dense sampling or precomputed scene-specific ray databases.

For sequential localization, FreeLoc integrates this online coarse-to-fine estimator into a probabilistic filtering pipeline. Rather than deriving dense observation likelihoods from an offline ray database, FreeLoc constructs them online from coarse-sampled candidates and refines their high-confidence regions using refined pose hypotheses.

This online construction bridges single-frame localization and temporal fusion without reintroducing a scene-specific database. On both tested indoor settings, FreeLoc consistently outperforms all compared methods. Compared with the strongest comparison method, it improves all recall measures by four point seven to eight point seven percentage points in one setting and one point four to eight point nine in the other.

For sequential localization, FreeLoc achieves the best success rate and overall tracking accuracy. Compared with the strongest comparison method, it improves the success rate by thirteen point five percentage points and reduces overall tracking error by eighty-two point eight percent, showing substantially more robust tracking over long trajectories.

A handheld phone can begin with almost no certainty about where it is inside a large building, then gradually narrow that uncertainty as it moves. Here, its estimated route comes together with the known route, supporting the system’s use in real indoor spaces.

The unified coarse-to-fine framework substantially reduces overall localization cost by eliminating the time-consuming scene-specific preprocessing stage. That means a new building can be approached without first spending time preparing a separate database for it.

FreeLoc treats the floorplan as a directly queryable geometric map rather than a dense scene-specific offline database, combining on-the-fly querying with pose refinement to turn plausible anchors into accurate continuous pose estimates. Experiments demonstrate accurate and efficient localization, as well as immediate deployability in new environments.

FreeLoc turns a floorplan into a map the robot can question on the spot, then sharpens rough guesses into continuous positions. That makes accurate indoor tracking more practical in unfamiliar buildings.