Learning IMU Bias with Diffusion Model

2025-05-17

AI Summary

IMU time-varying random biases—driven by temperature, vibration, and other environmental factors—are inherently stochastic and dynamic, rendering deterministic modeling inadequate. Existing deep learning approaches predominantly formulate bias estimation as a regression task, neglecting the intrinsic randomness and temporal dependencies of the bias process. Method: This work introduces, for the first time, a generative framework that models IMU bias as a conditional probability distribution. Leveraging a conditional diffusion model, we perform end-to-end learning of the mapping from raw IMU time-series measurements to the full bias distribution. Our approach explicitly encodes stochastic evolution dynamics and physical constraints (e.g., thermal drift patterns). Contribution/Results: The proposed method achieves significantly improved prediction accuracy and enhanced physical consistency compared to deterministic baselines. Experiments on real-world data demonstrate superior fidelity in capturing temperature- and vibration-driven bias dynamics, thereby overcoming fundamental limitations of conventional regression-based paradigms.

Technology Category

• Intelligent Robots: State Estimation
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

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.

Problem

• Modeling time-varying stochastic IMU bias accurately
• Addressing bias complexity from factors like temperature
• Predicting bias as probabilistic distribution, not regression

Innovation

• Model IMU bias as probabilistic distribution
• Use conditional diffusion model
• Improve prediction accuracy