IMUNavigator: An iOS High-Precision Inertial Navigation System Integrating ARKit, PDR, and Neural Networks

The IMUNavigator introduced in this article is a high-precision sensor fusion navigation app for the iOS platform. It deeply integrates ARKit Visual-Inertial Odometry (VIO), Pedestrian Dead Reckoning (PDR), and RoNIN neural network inertial navigation technology to achieve GPS-free indoor high-precision positioning and trajectory tracking. The project is developed and maintained by NalaniFlynns, open-sourced on GitHub (link: https://github.com/NalaniFlynns/IMUNavigator), released on June 9, 2026, uses the MIT License, supports iOS 16.0+, and is developed in Swift 5.9.

In indoor environments where GPS signals are unavailable (such as shopping malls, underground parking lots, tunnels, etc.), traditional Inertial Navigation Systems (INS) have cumulative drift errors, leading to a rapid decline in positioning accuracy. IMUNavigator uses a multi-sensor fusion solution, combining VIO, PDR, and neural network technologies to build an inertial navigation system that maintains high precision even in pure IMU mode, providing solutions for scenarios like indoor navigation and AR applications.

IMUNavigator uses a modular cascaded sensor fusion engine, with core components including:

1. ARKit VIO: Fuses camera and IMU data to provide centimeter-level relative positioning accuracy, relying on visual features;
2. PDR: A pure inertial method based on gait detection and step length estimation, which can work in dark environments but has cumulative errors;
3. RoNIN Neural Network: End-to-end learning of displacement patterns from raw IMU data, with strong robustness, deployed via CoreML;
4. ZUPT: Detects static states to suppress cumulative drift;
5. Spatial Alignment Lab: Provides automatic/manual calibration functions to correct heading deviations.

The fusion strategy uses cascaded priority: prefer ARKit VIO (when visual conditions are good) → degrade to PDR + RoNIN (when the camera is unavailable). Both modes combine ZUPT to suppress drift. Drift compensation mechanisms include automatic yaw alignment, manual calibration, and independent axis damping to ensure positioning continuity and stability.

IMUNavigator can be applied to:

• Indoor Navigation: Large shopping malls, airports, and other GPS-weak environments;
• AR Applications: Provides stable 6-degree-of-freedom pose estimation;
• Motion Tracking: Trajectory recording in GPS-weak areas such as dense forests and canyons;
• Emergency Positioning: Dark and GPS-unavailable scenarios like disaster rescue.

Project innovations include:

1. Multi-modal Cascaded Fusion: Intelligently selects the optimal positioning source to adapt to different environments;
2. On-device Neural Network Deployment: RoNIN implements offline inference via CoreML, protecting privacy and ensuring real-time performance;
3. Comprehensive Calibration Mechanism: Combines automatic and manual calibration, lowering the threshold for use while supporting fine adjustments.

Development technology stack:

Technology	Purpose
Swift5.9	Core language
SwiftUI	UI framework
Canvas	Trajectory rendering
CoreMotion	IMU data collection
ARKit	VIO
CoreML	RoNIN inference
Deployment steps: Clone the repository → Open with Xcode → Select iOS physical device → Compile and run (physical device required for sensor functionality testing).

IMUNavigator demonstrates the organic combination of computer vision, traditional inertial navigation, and deep learning, solving the limitations of a single technology. It is an excellent example for iOS developers to learn the integration of ARKit, CoreMotion, etc., with clear and extensible code. The project represents the direction of mobile navigation: multi-sensor fusion and on-device intelligence. The MIT license allows free use and modification, facilitating community innovation.