LBA-CBF: Rapidly Adaptive Safety Filters via Parallel Dynamics Inference

This project explores the implementation and evaluation of Look-Back Adaptive Control Barrier Functions (LBA-CBF) to address the challenges of safety-critical control under abrupt, unmeasured regime changes. Traditional control barrier functions rely heavily on accurate nominal dynamics models, which can fail during sudden environmental transitions or payload variations. The LBA-CBF framework mitigates this vulnerability by maintaining a finite bank of candidate dynamics models and ranking them based on their prediction error over a short, sliding look-back window. By enforcing high-order control barrier function conditions against all candidate models within a specified tolerance of the best-performing model, the system achieves a robust balance between best-fit adaptation and full-bank filtering. The methodology supports

nonlinear parameter dependencies without requiring continuous parameter estimators or explicit switching models. This research project focuses on simulating the LBA-CBF algorithm in highly dynamic environments, such as quadrotors experiencing wind reversals or vehicles encountering varying tire-road friction. The evaluation analyzes safety preservation, computational efficiency of parallel dynamics evaluation, and convergence properties compared to conventional robust and adaptive baselines, providing a structured framework for implementing resilient safety filters in real-time embedded systems.

Message: Hi FE, I need help with “LBA-CBF: Rapidly Adaptive Safety Filters via Parallel Dynamics Inference” in “Electronics & Communication Engineering”