<aside> 🚘
Research project on safe lane keeping using Hamilton–Jacobi reachability and Control Barrier Functions, with a primary focus on building and validating a 2D kinematic-model-based safety filter before extending the framework to higher-dimensional vehicle dynamics.
</aside>
Kinematic Single-Track Vehicle Model with bounding box
Illustration of vehicle configurations where one of the corners of the bounding box are touching the lane boundaries
This project studies how to guarantee safety for lane keeping without giving up smooth, reference-following control behavior.
The core question is:
How can we use HJ reachability analysis as a ground-truth safety certificate and then convert that insight into a real-time CBF-based safety filter that remains valid under realistic input constraints?
A key point in the actual project timeline is that the work did not proceed as a balanced 2D / 4D study. After some early exploration of 4D dynamic modeling, the project direction was intentionally narrowed so that the 2D kinematic model would be understood and validated first, in a rigorous way.
As a result, the main implemented and validated pipeline in this project is centered on:
The 4D dynamic model appears only as a preliminary exploratory direction and is treated at the end of this page as part of future work, rather than as a completed result.
Overview of the proposed pipeline (δ = 10.0°)
Classical CBFs are attractive because they are implementable online and can provide real-time safety guarantees through small optimization problems. However, in practice they often suffer from three major limitations: