M-CTX Accelerates Spatial Context Retrieval for Trajectory Analytics

Modern trajectory predictors increasingly rely on external spatial context, such as map geometry, signed distance fields (SDFs), and nearby moving agents, to improve prediction quality. However, generating this context for every training anchor has become a significant systems bottleneck. For instance, in a typical maritime AIS pipeline, spatial context construction can take weeks of CPU time, far exceeding the cost of the predictor itself. To address this, M-CTX has been developed as an exact and scalable spatial context-retrieval framework for trajectory analytics. M-CTX redefines context construction as a spatial database workload, where data is ingested once and queried many times. The framework replaces three computationally intensive brute-force stages—OSM range retrieval, SDF computation, and moving-vessel neighbor lookup—with composable, index-backed operators. Its learned range-index backend, BR-LZ, provides recall-complete MBR-overlap range retrieval and significantly reduces candidate amplification. M-CTX has demonstrated a 226x end-to-end speed-up in context construction, reducing it from days to hours, while reproducing reference context exactly. An optional storage mode also offers substantial SDF context compression.