RIO: Rotation-equivariance supervised learning of robust inertial odometry
This addresses the problem of data efficiency and generalization for inertial odometry in robotics or navigation, representing an incremental improvement through novel self-supervision and adaptation techniques.
The paper tackles the problem of training robust inertial odometry models by introducing rotation-equivariance as a self-supervisor, which reduces reliance on labeled data. The method achieves on-par performance with only 30% of training data and improves performance by over 25% in several scenarios using adaptive test-time training.
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.