Learnable Spatio-Temporal Map Embeddings for Deep Inertial Localization

Dennis Melamed
Karnik Ram
Vivek Roy
Kris Kitani

Robotics Institute, Carnegie Mellon University

Teaser figure

Our proposed method utilizes learnable spatio-temporal map priors to reduce drift in inertial odometry.


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.


Paper thumbnail.

Learnable Spatio-Temporal Map Embeddings for Deep Inertial Localization

Dennis Melamed, Karnik Ram, Vivek Roy, Kris Kitani

  title={Learnable Spatio-Temporal Map Embeddings for Deep Inertial Localization},
  author={Melamed, Dennis and Ram, Karnik and Roy, Vivek and Kitani, Kris},
  booktitle={2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},



Presentation Video



System overview figure


This work was supported in part by Meta Project Aria.
Webpage template.