Learnable Spatio-Temporal Map Embeddings for Deep Inertial Localization

TL;DR

This paper introduces a learnable spatio-temporal map embedding approach that significantly improves inertial localization accuracy by providing a data-driven prior, outperforming traditional hand-defined methods and matching beacon-based positioning.

Contribution

The authors propose a novel data-driven map prior combining spatial and temporal embeddings, enhancing inertial localization robustness and accuracy across different odometry sources.

Findings

49% improvement in inertial-only localization accuracy

Matches performance of absolute positioning with Bluetooth beacons

Effective with both inertial and wheel encoder odometry

Abstract

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.