Taylor-Lagrange Control for Safety-Critical Systems

Wei Xiao, Anni LiarXiv 2025 · DOI · arXiv

Introduction

Ensuring safety and stability in nonlinear systems is a cornerstone of modern robotics and autonomous systems. This paper introduces Taylor-Lagrange Control (TLC), a novel framework that leverages Taylor’s theorem with the Lagrange remainder to provide rigorous safety guarantees. Unlike traditional Control Barrier Functions (CBFs) that often provide only sufficient conditions for safety, TLC offers a necessary and sufficient condition, significantly reducing conservativeness in high-order systems.

The Challenge of Safety in Nonlinear Systems

In safety-critical applications like self-driving cars or industrial drones, controllers must guarantee that the system never enters an "unsafe" state. The current industry standard, Control Barrier Functions (CBFs), faces two major hurdles:

  1. Conservativeness: CBFs often provide "sufficient but not necessary" conditions. This means they might block a perfectly safe maneuver because the mathematical "buffer" they create is too restrictive.
  2. High Relative Degree: For many systems, the control input (like acceleration) doesn't immediately affect the safety constraint (like distance). Differentiating the safety function multiple times to find the control input leads to High-Order CBFs (HOCBFs), which are mathematically complex and often limit performance by requiring the system to stay within an intersection of multiple safe sets.

Methodology: Taylor-Lagrange Control

The authors propose TLC as a more flexible alternative. Instead of looking only at the immediate derivative of a safety function, TLC expands the safety function over time using a Taylor series.

  • Lagrange Remainder: By utilizing the Lagrange form of the remainder, the control input appears explicitly in the expansion at an order equal to the system's relative degree. This accounts for the system's evolution over a future interval $[t_0, t]$.
  • Quadratic Programming (QP): To ensure practical implementation, safety constraints are reformulated into a sequence of QPs that can be solved in real-time.
  • Event-Triggered Mechanism: To address the risks of discrete-time execution (Zero-Order Hold), the authors introduce an event-triggered mechanism. This recalculates the control law only when the system approaches a pre-defined state boundary, ensuring continuous-time safety without excessive computation.

Theoretical Insights and the Complex Domain

One of the most striking findings of this research is how TLC relates to and extends existing frameworks. While standard HOCBFs require real-valued "class $\mathcal{K}$" functions, the equivalent TLC formulation can involve imaginary numbers, effectively extending HOCBFs into the complex domain.

Comparison of TLC and CBF frameworks.

As illustrated in the figure above, TLC provides a broader, necessary and sufficient condition for safety. This allows the controller to explore safer trajectories that a standard HOCBF would have pruned, providing more freedom for the system to move while maintaining strict safety boundaries.

Experimental Validation

The TLC method was evaluated using two high-fidelity simulation benchmarks:

  1. Adaptive Cruise Control (ACC): In a car-following scenario with a 1-second sampling time, event-triggered TLC strictly enforced the safety boundary ($h(x) \geq 0$), whereas standard HOCBFs and time-driven TLC allowed for slight safety violations due to the large gap between updates.
  2. 2D Robot Navigation: In a unicycle model navigating obstacles, event-triggered TLC achieved collision-free movement with significantly fewer QP solves than time-driven HOCBFs. Furthermore, TLC generated smoother, less "twitchy" trajectories compared to the more conservative HOCBF approach.

Limitations and Future Work

Despite its advantages, TLC relies on identifying a value $\xi$ within the Lagrange remainder that exists in a future time interval. Analytically identifying the exact $\xi$ for highly non-linear systems remains difficult. Current implementations use Zero-Order Hold approximations and robust bounds, but future work aims to develop more formal bounding methods. Additionally, the authors plan to investigate the robustness of TLC in systems with unmodeled dynamics and integrate the framework into receding horizon control (MPC-like) architectures.

Summary updated

BibTeX
@misc{xiao2025taylor,
  author = {Wei Xiao and Anni Li},
  title = {{Taylor-Lagrange Control for Safety-Critical Systems}},
  year = {2025},
  eprint = {2512.11999},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.11999},
}
Figure