Learnable Spatio-Temporal Map Embeddings for Deep Inertial Localization
This work addresses the robustness issue in map utilization for indoor localization systems, which is incremental as it builds on existing methods but shows strong specific gains.
The paper tackled the problem of indoor localization by proposing a data-driven prior that combines learned spatial map and temporal odometry embeddings to reduce odometry drift, resulting in a 49% improvement in inertial-only localization accuracy and matching the performance of absolute positioning with bluetooth beacons.
Indoor localization systems often fuse inertial odometry with map information via hand-defined methods to reduce odometry drift, but such methods are sensitive to noise and struggle to generalize across odometry sources. To address the robustness problem in map utilization, we propose a data-driven prior on possible user locations in a map by combining learned spatial map embeddings and temporal odometry embeddings. Our prior learns to encode which map regions are feasible locations for a user more accurately than previous hand-defined methods. This prior leads to a 49% improvement in inertial-only localization accuracy when used in a particle filter. This result is significant, as it shows that our relative positioning method can match the performance of absolute positioning using bluetooth beacons. To show the generalizability of our method, we also show similar improvements using wheel encoder odometry.