⚡ electrical engineering

From Crash Prone Gap Following to Robust Safety via Nonlinear Control Barrier Functions

This paper demonstrates that implementing a speed-aware nonlinear Control Barrier Function on an autonomous vehicle effectively prevents adversarial overtaking-induced crashes and ensures track invariance, whereas standard and linearly augmented gap-following policies fail under similar conditions.

Original authors: Hossein Talebi Rostami, Farzad TalebiRostami

Published 2026-09-09

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Hossein Talebi Rostami, Farzad TalebiRostami

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The road is a place of constant negotiation. Even in the most orderly traffic, drivers must constantly judge distance, speed, and the unpredictable movements of others. For autonomous vehicles, which must make these judgments without human intuition, the challenge is even greater. A car that simply follows a path or reacts to the space immediately in front of it can find itself in trouble when another vehicle acts aggressively or erratically. To solve this, researchers are developing safety systems that do more than just react; they anticipate danger by understanding how fast a car is moving and how quickly it is changing direction. The goal is to create a digital shield that keeps a vehicle within its lane, even when pushed to the edge by a chaotic environment.

In a recent study, researchers put this idea to the test using small, real-world autonomous cars on a straight, narrow track. They set up a scenario where one car, acting as a primary vehicle, would perform disruptive, zig-zagging maneuvers to block a second car, the pursuer, from passing. The researchers wanted to see if standard navigation methods could keep the pursuer safe. They found that the most common approach, which simply steers toward the widest available gap between obstacles, failed spectacularly. When the lead car wove back and forth, the pursuer, trying to follow the shifting gap, was driven into the walls of the corridor. Even a more cautious version of this method, which aimed for the center of the gap rather than the edge, could not prevent the crash. The problem was that these methods looked at where the car was, but not how fast it was moving sideways. A car moving slowly toward a wall has time to turn, but a car moving fast toward a wall needs to turn much sooner.

To fix this, the team designed a new safety rule that combined position with speed. They created a mathematical boundary that tightened as the car's sideways speed increased. This meant that if the car began to drift toward the wall while moving quickly, the safety system would force an earlier, stronger correction back toward the center of the lane. They tested this new rule alongside the old methods. When the lead car performed its aggressive zig-zag, the pursuer equipped with the new speed-aware rule stayed safely within the track every time. It made small, smooth adjustments to stay centered, avoiding the violent swings that caused the other cars to crash. The researchers measured the performance on a track one meter wide and ten meters long, using a high-speed camera system to track the cars' movements with extreme precision. The results showed that while the traditional methods led to collisions in multiple trials, the new approach kept the vehicle safe without slowing it down significantly.

The study highlights a critical lesson for the future of self-driving cars: safety cannot be guaranteed by looking only at where a vehicle is. It must also account for how the vehicle is moving. The researchers demonstrated that by adding a simple awareness of lateral speed to their safety calculations, they could prevent a vehicle from leaving its lane even under the most stressful conditions. This work was conducted on the Agilex Limo platform, a small autonomous vehicle, in a controlled environment designed to mimic the tight constraints of a real-world overtaking maneuver. The findings suggest that for autonomous systems to be truly robust, their safety logic must evolve from static boundaries to dynamic ones that change based on the vehicle's momentum. By doing so, engineers can build systems that do not just avoid crashes in theory, but survive the chaotic reality of the road.

Technical Summary: From Crash-Prone Gap Following to Robust Safety via Nonlinear Control Barrier Functions

Problem Statement
The paper addresses the safety limitations of standard gap-following navigation methods, specifically the Follow-the-Gap (FTG) algorithm, when autonomous vehicles (AVs) operate in adversarial overtaking scenarios. While FTG and its variants are efficient for mapless navigation, they often fail under rapidly shifting gap configurations or aggressive maneuvers by a lead vehicle. These failures manifest as "border crashes," where the pursuer vehicle leaves the track and collides with the corridor walls. The core challenge is distinguishing between controllers that appear safe under nominal conditions but fail under adversarial stress (such as linear Control Barrier Functions) and those that provide genuine robustness.

Methodology
The study utilizes the Agilex Limo autonomous vehicle platform within a straight corridor environment (1m width, 10m length) monitored by a Vicon motion capture system. The research compares four distinct scenarios involving a primary "lead" vehicle executing adversarial zig-zag maneuvers and a secondary "pursuer" vehicle attempting to overtake.

System Modeling: The vehicle dynamics are modeled using a kinematic unicycle model with first-order velocity tracking. The track safety is defined by lateral position constraints ( $y \in [-W/2, W/2]$ ).
Controller Design:
- Baseline Controllers: Standard FTG (steering toward the outer edge of the widest gap) and a Conservative FTG variant (steering toward the gap center with tighter rate limits).
- Safety Filters: The authors implement Control Barrier Functions (CBFs) to enforce track invariance. They compare a linear CBF (based purely on position) against a proposed nonlinear, velocity-aware CBF.
- Nonlinear CBF Formulation: The proposed function, $h(y, \dot{y}) = \frac{W^2}{4} - (1+\gamma)y^2 - \beta\sqrt{\dot{y}^2 + \epsilon}$ , integrates lateral position ( $y$ ) and lateral velocity ( $\dot{y}$ ). This formulation includes a centering term and a velocity-dependent tightening term that reduces the admissible band as lateral speed increases, forcing earlier corrective action near boundaries.
Adversarial Setup: The lead vehicle is programmed with a "zig-zag" maneuver involving rapid lateral oscillations (micro-oscillations) and boundary-triggered corrections (macro-oscillations) to disrupt the pursuer's gap detection.
Implementation: Safety constraints are enforced in real-time via a Quadratic Program (QP) that minimizes deviation from a nominal steering command while satisfying the CBF inequality $\dot{h} + \kappa h \geq 0$ .

Key Contributions
The paper makes four primary contributions:

Adversarial Maneuver Design: The creation of a repeatable zig-zag maneuver for the lead vehicle, characterized by micro-oscillations from control inputs and macro-oscillations resulting from boundary-triggered CBF corrections.
Nonlinear CBF Formulation: The development of a velocity-aware, zeroing CBF for lateral track containment. This defines a velocity-dependent safe set that adds robustness against the aggressive lateral excursions caused by adversarial driving.
Integration: The successful integration of the nonlinear CBF with FTG controllers to create safer overtaking strategies.
Experimental Validation: Empirical demonstration on the Agilex Limo platform showing that while FTG and linear CBFs fail under adversarial conditions, the nonlinear CBF maintains safety with minimal performance penalty.

Experimental Results
The authors evaluated four scenarios:

Scenario 1 (Single Vehicle): The nonlinear CBF successfully maintained track invariance for a single vehicle at maximum speed, correcting small excursions near boundaries.
Scenario 2 (FTG vs. Adversarial Lead): The standard FTG pursuer consistently crashed into the flexible duct borders due to rapid gap switching and aggressive steering.
Scenario 3 (Conservative FTG vs. Adversarial Lead): Despite reduced lateral excursions and smoother turns, the conservative variant still failed to prevent crashes in all seven trials, as the repeated gap reconfiguration overwhelmed the controller.
Scenario 4 (FTG with Linear vs. Nonlinear CBF):
- Linear CBF: Failed to reliably capture unsafe states. In 1 out of 5 trials with varying initial conditions, the pursuer drifted outward and crashed. The linear constraint lacked the necessary centering effect to prevent boundary violations when actuation limits were reached.
- Nonlinear CBF: In all 10 experiments, the pursuer remained within the lane. The velocity-aware tightening of the constraint prompted earlier inward steering, successfully preventing border-crossing crashes.

Significance and Claims
The paper claims that standard gap-following methods, even when augmented with conservative margins or linear safety filters, are insufficient for adversarial overtaking scenarios. The primary significance lies in demonstrating that nonlinearity and velocity awareness are critical for robust safety. The proposed nonlinear CBF provides a "more robust notion of safety" by dynamically adjusting the safe set based on lateral velocity, thereby preventing the vehicle from entering states where the controller cannot recover before a crash occurs. The authors conclude that their approach preserves safety while imposing only a minimal penalty on lap time, offering a principled framework for real-time safety in complex, unpredictable driving environments.

The authors note that future work will extend the framework to include tire slip, sensing latency, curved tracks, and intent inference, but the current study is strictly limited to the kinematic model and the specific adversarial overtaking scenarios tested.

Technical Summary: From Crash-Prone Gap Following to Robust Safety via Nonlinear Control Barrier Functions

More like this