UMLoc: Uncertainty-Aware Map-Constrained Inertial Localization with Quantified Bounds

Area of Science:

• Robotics
• Sensor Fusion
• Indoor Navigation

Background:

• Inertial localization using Inertial Measurement Units (IMUs) is crucial for GPS-denied environments but suffers from drift due to noise and sensor biases.
• Existing methods struggle to accurately model sensor uncertainty and integrate map information effectively for robust positioning.

Purpose of the Study:

• To introduce Uncertainty-aware Map-constrained Inertial Localization (UMLoc), an end-to-end framework for drift-resilient indoor positioning.
• To jointly model IMU uncertainty and map constraints for enhanced localization accuracy and reliability.

Main Methods:

• UMLoc employs a Long Short-Term Memory (LSTM) quantile regressor to estimate localization uncertainty via prediction intervals.
• A Conditioned Generative Adversarial Network (CGAN) with cross-attention fuses IMU data with floor-plan maps for geometrically feasible trajectory generation.
• Both modules are trained jointly, enabling uncertainty propagation for improved trajectory estimation.

Main Results:

• The proposed UMLoc framework achieved a mean drift ratio of 5.9% over a 70m travel distance.
• An average Absolute Trajectory Error (ATE) of 1.36m was recorded, demonstrating high positioning accuracy.
• The system successfully maintained calibrated prediction bounds, indicating reliable uncertainty estimation.

Conclusions:

• UMLoc offers a robust solution for drift-resilient inertial localization in indoor environments.
• The joint modeling of IMU uncertainty and map constraints significantly improves trajectory generation accuracy.
• This framework provides a valuable tool for applications requiring precise indoor positioning without GPS.