Learned IMU Bias Prediction for Invariant Visual Inertial Odometry

2025-05-10

AI Summary

In unknown environments, IMU bias disrupts the Lie-group symmetry of visual-inertial odometry (VIO), degrading its convergence and robustness for autonomous mobile robots. Method: We propose a decoupled IMU bias prediction framework that treats bias estimation as an external learning task—rather than embedding it in the filter state—thereby rigorously preserving the Lie-group invariance of the VIO system. Our approach integrates a temporal deep neural network to model dynamic IMU bias with an invariant multi-state constraint Kalman filter (Invariant MSCKF), enabling seamless switching between visual-inertial and pure-inertial navigation modes. Results: Real-world experiments demonstrate significantly improved pose estimation robustness during prolonged visual degradation: pure-inertial navigation drift is reduced by 42%, and convergence speed increases by 3.1× compared to conventional methods.

Technology Category

• Intelligent Robots: State Estimation
• Computer Vision: Bias, Fairness & Privacy
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Autonomous mobile robots operating in novel environments depend critically on accurate state estimation, often utilizing visual and inertial measurements. Recent work has shown that an invariant formulation of the extended Kalman filter improves the convergence and robustness of visual-inertial odometry by utilizing the Lie group structure of a robot's position, velocity, and orientation states. However, inertial sensors also require measurement bias estimation, yet introducing the bias in the filter state breaks the Lie group symmetry. In this paper, we design a neural network to predict the bias of an inertial measurement unit (IMU) from a sequence of previous IMU measurements. This allows us to use an invariant filter for visual inertial odometry, relying on the learned bias prediction rather than introducing the bias in the filter state. We demonstrate that an invariant multi-state constraint Kalman filter (MSCKF) with learned bias predictions achieves robust visual-inertial odometry in real experiments, even when visual information is unavailable for extended periods and the system needs to rely solely on IMU measurements.

Problem

• Predicting IMU bias to maintain Lie group symmetry
• Improving visual-inertial odometry robustness with learned bias
• Enabling accurate state estimation during visual data loss

Innovation

• Neural network predicts IMU bias from measurements
• Invariant filter avoids bias in state symmetry
• MSCKF with learned bias ensures robust odometry