RIO: Rotation-equivariance supervised learning of robust inertial odometry

TL;DR

This paper presents RIO, a rotation-equivariance supervised learning method for inertial odometry that reduces data requirements and enhances generalization through adaptive test-time training, achieving high accuracy with less labeled data.

Contribution

Introduces rotation-equivariance as a self-supervised signal for inertial odometry and proposes adaptive test-time training to improve model robustness and generalization.

Findings

RIO trained with 30% data matches full-data performance.

Adaptive TTT improves accuracy by over 25%.

Self-supervision reduces reliance on labeled data.

Abstract

This paper introduces rotation-equivariance as a self-supervisor to train inertial odometry models. We demonstrate that the self-supervised scheme provides a powerful supervisory signal at training phase as well as at inference stage. It reduces the reliance on massive amounts of labeled data for training a robust model and makes it possible to update the model using various unlabeled data. Further, we propose adaptive Test-Time Training (TTT) based on uncertainty estimations in order to enhance the generalizability of the inertial odometry to various unseen data. We show in experiments that the Rotation-equivariance-supervised Inertial Odometry (RIO) trained with 30% data achieves on par performance with a model trained with the whole database. Adaptive TTT improves models performance in all cases and makes more than 25% improvements under several scenarios.