TODO.md

TODO - Project Enhancement Plan

This document outlines key features to implement to make the EKFCore project more robust.

---

High-Impact Features

• Implement an Outlier Rejection Mechanism

  • Calculate the Mahalanobis distance for each new measurement's innovation.
  • Compare the Mahalanobis distance to a threshold based on a chi-squared ($\chi^2$) distribution.
  • If the measurement is an outlier, discard it and skip the update step.

• Demonstrate Multi-Sensor Fusion

  • Create a single, generic update method that can handle different measurement types.
  • Inside the update method, apply the correct measurement model ($H$) and noise covariance ($R$) based on the incoming sensor data.
  • Verify that the filter can seamlessly switch between fusing different sensor data streams.

---

Validation and Analysis

• Add Consistency Checks ($\chi^2$ Tests)

  • Calculate the Normalized Innovation Squared (NIS) for each measurement.
  • Plot the NIS values over time and compare them to the theoretical $\chi^2$ distribution.
  • Use these plots to identify if the filter's covariance matrices ($Q$ and $R$) are well-tuned.

• Plot Covariance Bounds

  • Generate plots that show the filter's estimated path with $\pm 1\sigma$ (standard deviation) bounds.
  • Overlay these bounds with the ground truth path to visually confirm that the filter is accurately capturing its own uncertainty.

---

Future Work & Additional Ideas

• Implement a UKF (Unscented Kalman Filter) for comparison with the EKF.
• Add Dynamic Covariance Tuning to allow Q and R to change based on context (e.g., higher noise in a GPS when in a tunnel).
• Integrate with a real-world dataset or a real robotics framework like ROS 2.
• Implement a Full SLAM Solution by extending the state to include landmarks.