A Kalman Filter implementation in Kotlin, designed with built-in support for debugging, calibration, visualization, and additional tools to enhance usability.
This repository is actively under development, and its API, architecture, or functionality may change frequently. Backward compatibility is not guaranteed, and breaking changes may occur without prior notice.
If you're using this project, following closely with development updates. Contributions and feedback are welcome!
A Kalman Filter is an algorithm that provides an optimal estimate of multidimensional variables in dynamic system by iteratively refining predictions based on noisy sensor data. It is widely used in robotics, tracking systems, navigation, and signal processing due to its efficiency in filtering and predicting system states.
Specific use case of smartphone, including:
- GPS Positioning & Navigation Problem: GPS signals are often noisy due to atmospheric interference, signal reflection, or loss in urban environments.
- Motion Tracking & Inertial Navigation (IMU Sensors) Problem: Smartphone accelerometers, gyroscopes, and magnetometers provide raw, noisy sensor readings.
- Camera Image Stabilization (OIS & EIS) Problem: Shaky hands cause unstable video recording and blurry photos.
- Touchscreen Smoothing & Gesture Recognition Problem: Raw touchscreen input contains noise, leading to shaky strokes in handwriting apps or inconsistent gestures.
- AR & VR of sensor reading to predict state of system.
Mobile Application: Apps like Ski Tracks, Strava, and Runkeeper use Kalman filtering to smooth motion tracking and estimate accurate speed and position.
- Streamlined Calibration – Tuning a Kalman Filter requires extensive trial and error. This repository aims to provide interactive and extendable tools for calibrating systems modeling.
- Native Kotlin Support – Unlike other implementations, this repository is built natively in Kotlin, offering seamless integration with the Kotlin ecosystem.
- Enhanced Modeling Flexibility – Traditional Kalman Filter implementations often assume fixed parameters per iteration. This implementation supports dynamic covariance updates, enabling more adaptable system modeling.
This graph was generated by linear_2d_calibration.ipynb, demonstrate example of how to calibrate a 2D linear system using artificial calibration datasets and a Kalman Filter. Below is a breakdown of the key elements and their significance:
Key Elements:
-
Red Line (Calibration_Hidden_State):Represents the true hidden state of the system, which acts as the ideal baseline for calibration.
-
Gray Line (KF_Measurement): Shows the noisy measurements of the system's state, reflecting the raw data collected during calibration.
-
Purple Line (KF_Estimate): Displays the Kalman Filter's estimated state, which progressively converges toward the true hidden state (red line), reducing noise.
What This Graph Demonstrates:
-
Noise in Measurements: The gray line represents noisy inputs commonly encountered in real-world systems. These fluctuations highlight the challenge of working with raw data.
-
Kalman Filter's Role: The purple line demonstrates how the Kalman Filter smooths the noisy data, providing a more accurate estimate of the system's state.
-
Calibration Accuracy: The red line serves as a benchmark, showing how closely the Kalman Filter's estimate aligns with the true state over time.
Takeaway: This graph highlights the effectiveness of the Kalman Filter in handling noisy data and improving system calibration, making it a valuable tool for applications requiring precise state estimation.
:core– The fundamental Kalman Filter implementation, designed with detailed documentation explaining its inner workings. This package offers maximum flexibility for system modeling and can be extended as needed. It will be packaged for distribution.:utility– Provides utility functions for testing, debugging, and calibrating both KalmanFilterCore and its application use cases:src- Contains examples demonstrating the usage of the Kalman Filter.
The foundation of the Kalman Filter relies on matrix and vector operations for multidimensional system modeling. This implementation leverages third-party libraries for efficient computation.
Since Kotlin lacks built-in matrix and vector operations, this project uses:
Apache Commons Math – A stable and well-supported mathematical library.
This folder contains calibrated examples of Kalman Filter for custom systems. It serves as both:
- The practical tutorial to learn the Kalman Filter with visualization.
- The debugging and interactive tool, allowing developers to refine system parameters for different application use cases.
Powered by Kotlin Notebook – Learn More.
This is one of the core features of the repository, providing interactive debugging, visualization, and rapid iteration.
- Each Notebook file should focus on a single, unique system to serve as a clear example for developers, making it easier to find the relevant use case.
- Notebook files should follow clean code practices and include clear documentation.
- Custom classes and components used in a Notebook should be unit tested, or have unit tests added in
src/test. - Notebook filenames should follow the convention:
{system_modeling}_calibration.ipynb(e.g.,linear_2d_calibration.ipynb).
This project is open to contributions! Whether you want to improve documentation, optimize performance, or expand features, feel free to open an issue or submit a pull request.
🚀 Happy coding!