Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning IMU Bias with Diffusion Model

Shenghao Zhou, Saimouli Katragadda, Guoquan Huang

TL;DR

The paper addresses the challenge of time-varying stochastic IMU bias in inertial-only odometry by reframing bias as a conditional probability distribution. It introduces a lightweight conditional diffusion model that uses an IMU-derived conditioning code to learn the IMU-conditioned bias distribution, enabling bias samples to correct IMU readings during integration. Empirical results on EuRoC show the diffusion-based approach outperforms regression and random-walk baselines, with comparable performance to indirect supervision methods while producing more faithful bias traces. The method demonstrates real-time viability on edge hardware, offering a practical probabilistic framework for improving IOO without relying on exteroceptive sensors. The work opens avenues for uncertainty-aware bias predictions and distribution-based bias utilization in motion estimation.

Abstract

Motion sensing and tracking with IMU data is essential for spatial intelligence, which however is challenging due to the presence of time-varying stochastic bias. IMU bias is affected by various factors such as temperature and vibration, making it highly complex and difficult to model analytically. Recent data-driven approaches using deep learning have shown promise in predicting bias from IMU readings. However, these methods often treat the task as a regression problem, overlooking the stochatic nature of bias. In contrast, we model bias, conditioned on IMU readings, as a probabilistic distribution and design a conditional diffusion model to approximate this distribution. Through this approach, we achieve improved performance and make predictions that align more closely with the known behavior of bias.

Learning IMU Bias with Diffusion Model

TL;DR

The paper addresses the challenge of time-varying stochastic IMU bias in inertial-only odometry by reframing bias as a conditional probability distribution. It introduces a lightweight conditional diffusion model that uses an IMU-derived conditioning code to learn the IMU-conditioned bias distribution, enabling bias samples to correct IMU readings during integration. Empirical results on EuRoC show the diffusion-based approach outperforms regression and random-walk baselines, with comparable performance to indirect supervision methods while producing more faithful bias traces. The method demonstrates real-time viability on edge hardware, offering a practical probabilistic framework for improving IOO without relying on exteroceptive sensors. The work opens avenues for uncertainty-aware bias predictions and distribution-based bias utilization in motion estimation.

Abstract

Motion sensing and tracking with IMU data is essential for spatial intelligence, which however is challenging due to the presence of time-varying stochastic bias. IMU bias is affected by various factors such as temperature and vibration, making it highly complex and difficult to model analytically. Recent data-driven approaches using deep learning have shown promise in predicting bias from IMU readings. However, these methods often treat the task as a regression problem, overlooking the stochatic nature of bias. In contrast, we model bias, conditioned on IMU readings, as a probabilistic distribution and design a conditional diffusion model to approximate this distribution. Through this approach, we achieve improved performance and make predictions that align more closely with the known behavior of bias.
Paper Structure (16 sections, 14 equations, 2 figures, 2 tables)

This paper contains 16 sections, 14 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Model-free Method
  4. Model-based Method
  5. Inertial-Only Odometry
  6. Inertial Navigation
  7. Modeling Bias
  8. Learning Bias for IOO
  9. Diffusion Model
  10. Model Design
  11. Implementation Details
  12. Acquire Bias Ground Truth Data
  13. Experimental Results
  14. Diffusion Model vs. Indirect Regression
  15. Timing on Embedded Device
  16. ...and 1 more sections

Figures (2)

  • Figure 1: System overview: our model consists of IMU encoder and denoiser network of the diffusion model. Conditional code $\mathbf{c}$ extracted by IMU encoder from IMU readings is passed to the denoiser network, to generates the bias with multiple diffusion steps. Bias is used to correct the IMU readings for integration, to get the motion estimation.
  • Figure 2: Bias prediction result for our model and AirIMU in an one-second window