💻 computer science

Two-dimensional time-to-collision measures for articulated vehicles: predicting sideswipe and rear-end collisions

This paper proposes refined two-dimensional time-to-collision measures for articulated vehicles that incorporate heading differences and constant-acceleration assumptions, demonstrating significantly improved detection of sideswipe collisions while maintaining effective rear-end collision prediction in simulated scenarios.

Original authors: Abhijeet Behera, Sogol Kharrazi, Erik Frisk, Maytheewat Aramrattana

Published 2026-08-17

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Abhijeet Behera, Sogol Kharrazi, Erik Frisk, Maytheewat Aramrattana

Original paper licensed under CC BY 4.0 (http://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

Imagine you are standing on a busy highway, watching the world zoom by. In the world of traffic safety, scientists use a special "crystal ball" called Time-to-Collision (TTC) to predict accidents before they happen. Think of TTC as a countdown timer that tells you how many seconds you have before your car hits the one in front of you. For decades, this timer has been incredibly good at spotting rear-end collisions—the kind where you bump into the back of a car because you were going too fast or didn't brake in time. It works by measuring the distance between cars and how fast they are closing that gap, kind of like a game of "chicken" played in one straight line.

But here's the catch: that old timer only looks straight ahead. It completely ignores what's happening on the sides. If a giant truck decides to swerve into your lane from the side, the old timer might say, "No problem, you won't hit the back of the truck!" while you are actually about to get clipped on the side. This is called a sideswipe collision. It's like trying to catch a ball thrown at you from the side while only looking forward; you'll miss it every time. This is a huge problem for articulated vehicles—those massive trucks made of a tractor pulling a long trailer (like a semi-truck). Because these trucks are so long and can bend in the middle, they take up a lot of space sideways. When they change lanes, they don't just move forward; they sweep a huge arc, creating a "sweeping" danger zone that old safety tools simply can't see.

This paper is about building a brand new, two-dimensional crystal ball that can see both forward and sideways, specifically designed for these giant, bending trucks. The researchers, working with a super-advanced video game simulator called CARLA, created a new way to calculate crash risks that accounts for the truck's length, its trailer's length, and the fact that the truck and trailer might be pointing in slightly different directions. They tested their new math against the old methods using 30 different crash scenarios, including tricky lane changes, tight turns, and roundabouts.

The results are a game-changer for spotting side-swipes. In their simulations, the old methods failed to identify the correct type of crash (sideswipe vs. rear-end) in half of the cases, often missing the danger entirely. The new method, however, correctly identified 14 out of 15 sideswipe collisions. It also got the timing of the crash much more accurate, reducing the prediction error by about 20% compared to the existing formulas. For rear-end collisions, the new method performed just as well as the old ones, proving it doesn't break what already works. The team also tested the system in complex situations like roundabouts and tight turns, where the truck's trailer swings out into the next lane. In these tricky spots, the old tools often said "no crash" when a crash was actually happening, while the new tools correctly predicted the sideswipe.

The paper introduces two specific versions of this new tool. The first, called TTCAV2D, assumes the truck and car are moving at a steady speed. The second, MTTCAV2D, is a bit smarter; it assumes the vehicles might be speeding up or slowing down (constant acceleration), which is more realistic for heavy trucks that take a while to change speed. While the "acceleration" version is theoretically more advanced, the simulations showed that for short, quick lane changes, both versions worked almost identically well. The researchers found that their new approach is fast enough to run in real-time on a computer, taking only a tiny fraction of a second to calculate, which is perfect for actual safety systems in vehicles.

Ultimately, this work suggests that we can't just look straight ahead to stay safe on the roads anymore, especially when sharing the road with giant, flexible trucks. By adding a "side view" to our safety calculations, we can catch those dangerous side-swipes before they happen. The study confirms that while the new tools are a significant improvement for detecting side collisions, they are equally reliable for the traditional rear-end crashes. The authors note that while their results are based on computer simulations, the math is solid and ready to be tested on real roads, potentially leading to warning systems that can save lives by telling drivers, "Hey, that truck isn't just in front of you; it's swinging into your lane!"

Technical Summary: Two-dimensional Time-to-Collision Measures for Articulated Vehicles

Problem Statement
Time-to-Collision (TTC) is a standard surrogate safety measure (SSM) primarily designed to predict rear-end collisions based on longitudinal motion. However, conventional TTC formulations fail to detect lateral collisions, such as sideswipes, which are governed by relative lateral displacement and trajectory overlap. While recent two-dimensional (2D) extensions of TTC have been proposed to address lateral risks for passenger cars, they rely on the assumption that interacting vehicles maintain identical and fixed headings. This assumption is invalid for articulated vehicles, such as tractor-semitrailers, where the tractor and semitrailer units can possess distinct and varying heading angles. Consequently, existing 2D TTC measures are inadequate for accurately assessing collision risks involving articulated vehicles, particularly in scenarios involving lane changes, intersections, and roundabouts where off-tracking and trailer swing create complex lateral interactions.

Methodology
The paper proposes a refined framework for 2D TTC that accounts for vehicle orientation and articulation dynamics. The methodology proceeds in three stages:

Refinement of 2D TTC for Non-Articulated Vehicles: The authors first extend the existing 2D TTC formulation (Guo et al., 2023) to accommodate differences in vehicle heading. By employing a coordinate transformation framework, the lead vehicle's dimensions (length and width) and velocity vectors are projected onto the follower vehicle's local coordinate system. This allows for the calculation of longitudinal and lateral time-to-collision ( $TTC_{0X}$ and $TTC_{0Y}$ ) that accounts for yaw angle differences ( $\psi$ ) between vehicles.
Development of Articulated Vehicle Measures: Two new measures are formulated specifically for tractor-semitrailers:
- $TTC^{AV}_{2D}$ : Assumes constant speed and constant heading for both the tractor and the semitrailer over the prediction horizon. The collision risk is evaluated by computing the 2D TTC between the car and the tractor ( $TTC_{2D,0}$ ) and between the car and the semitrailer ( $TTC_{2D,1}$ ). The semitrailer's trajectory is derived analytically based on the tractor's trajectory and the articulation constraint. The final measure is the minimum of these two values.
- $MTTC^{AV}_{2D}$ (Modified $TTC^{AV}_{2D}$ ): Extends the previous measure by replacing the constant speed assumption with a constant acceleration assumption, analogous to the Modified TTC (MTTC) concept. This formulation incorporates acceleration terms into the kinematic equations to better capture dynamic changes in velocity, while maintaining the constant heading assumption within the prediction horizon.
Simulation and Evaluation: The proposed measures were evaluated using the CARLA high-fidelity simulation environment. A high-fidelity tractor-semitrailer model was implemented with varying trailer lengths (10m, 13m, 15m). The evaluation utilized 30 randomly generated "cut-in" scenarios, as well as additional scenarios involving tight turns and roundabouts. The performance was benchmarked against conventional TTC and the adjusted existing 2D TTC (where the existing formulation was applied separately to the tractor and trailer without accounting for articulation dynamics).

Key Contributions

Coordinate Transformation Framework: An extension of the 2D TTC formulation to handle non-aligned vehicle headings through rotation matrices, improving accuracy for non-articulated vehicles in diverse traffic environments.
$TTC^{AV}_{2D}$ : The first 2D TTC measure explicitly formulated for articulated vehicles, capable of predicting both rear-end and sideswipe collisions by modeling the kinematic coupling between the tractor and semitrailer.
$MTTC^{AV}_{2D}$ : A constant-acceleration extension of the articulated vehicle measure, providing a two-dimensional analogue to the Modified TTC.
Comprehensive Validation: A comparative analysis demonstrating the efficacy of these measures in detecting sideswipe collisions, which existing formulations frequently miss or misclassify.

Results
The evaluation across 30 cut-in scenarios yielded the following findings:

Sideswipe Detection: The proposed measures significantly outperformed existing formulations in identifying sideswipe collisions. They correctly identified 14 out of 15 sideswipe scenarios, compared to only 7 out of 15 identified by the existing 2D TTC formulation.
Prediction Accuracy: For sideswipe collisions, the mean prediction error of the proposed measures was reduced by approximately 20% compared to the existing formulation. The existing formulation tended to overestimate the time-to-collision for lateral conflicts.
Rear-end Detection: In rear-end collision scenarios, the proposed measures maintained a performance level comparable to existing measures, correctly identifying the collision type and timing.
Complex Scenarios: In roundabout and tight turn scenarios, the proposed measures successfully predicted sideswipe collisions caused by off-tracking. In contrast, the conventional TTC failed to detect any collision in the roundabout scenario, and the existing 2D TTC overestimated the collision time.
Consistency: Statistical analysis using bootstrap resampling indicated that the proposed measures provide more consistent predictions (narrower confidence intervals) than the existing formulation, particularly at prediction horizons of 1.38 seconds and 1 second prior to collision.

Significance and Claims
The paper claims that the proposed measures address a critical gap in surrogate safety assessment for articulated vehicles. By explicitly modeling the articulation constraints and vehicle orientation, the measures enable the accurate prediction of lateral collision risks that are invisible to conventional longitudinal TTC. The authors state that these measures facilitate more accurate safety assessments in transportation simulations and provide a foundation for evaluating emerging vehicle technologies and operational strategies for heavy freight. The work supports the development of driver warning systems that can distinguish between rear-end and sideswipe risks, potentially leading to more intuitive and effective collision avoidance interventions. The authors note that while the measures were evaluated in simulation, they are designed to be compatible with real-world sensor data (e.g., radar, LiDAR, V2X) required for state estimation, though practical deployment requires further research into handling sensing uncertainties and estimation errors.

More like this