Denoising IMU Gyroscopes with Deep Learning for Open-Loop Attitude\n Estimation

Abstract

This paper proposes a learning method for denoising gyroscopes of Inertial\nMeasurement Units (IMUs) using ground truth data, and estimating in real time\nthe orientation (attitude) of a robot in dead reckoning. The obtained algorithm\noutperforms the state-of-the-art on the (unseen) test sequences. The obtained\nperformances are achieved thanks to a well-chosen model, a proper loss function\nfor orientation increments, and through the identification of key points when\ntraining with high-frequency inertial data. Our approach builds upon a neural\nnetwork based on dilated convolutions, without requiring any recurrent neural\nnetwork. We demonstrate how efficient our strategy is for 3D attitude\nestimation on the EuRoC and TUM-VI datasets. Interestingly, we observe our dead\nreckoning algorithm manages to beat top-ranked visual-inertial odometry systems\nin terms of attitude estimation although it does not use vision sensors. We\nbelieve this paper offers new perspectives for visual-inertial localization and\nconstitutes a step toward more efficient learning methods involving IMUs. Our\nopen-source implementation is available at\nhttps://github.com/mbrossar/denoise-imu-gyro.\n