Traffic safety is often assessed at a single moment: how close two vehicles are, how fast they are moving or how quickly a collision could occur. A new study argues that this approach may miss an important part of the picture.

Published in Transportation Research Part C: Emerging Technologies, the research introduces the concept of “safety hysteresis” — a pattern in which traffic risk does not rise and fall in exactly the same way during a disturbance. Instead, the safety state depends partly on what happened moments earlier.

What Safety Hysteresis Means

Hysteresis describes a system whose current condition depends on its recent history. In traffic, the researchers found that safety could deteriorate as vehicles slowed and then recover along a different path as traffic accelerated again.

At the level of individual vehicle pairs, this appeared as counterclockwise loops when spacing was compared with inverse time-to-collision, a safety measure that rises as vehicles approach a potentially dangerous situation.

At the platoon level, involving groups of vehicles, the pattern appeared as clockwise loops when traffic density was compared with average inverse time-to-collision.

The important point is that identical traffic conditions can be associated with different levels of risk depending on whether congestion is building or easing.

How The Researchers Studied It

The team, led by Peilin Zhao with Yiik Diew Wong and Feng Zhu, analysed several large-scale vehicle-trajectory datasets.

These included NGSIM and Waymo data representing naturalistic traffic, OpenACC data from controlled adaptive-cruise-control experiments, and additional simulations designed to isolate specific causes of the effect.

To compare the strength of the phenomenon across different situations, the researchers proposed a new measure called Safety Hysteresis Intensity, or HI. It is based on the normalised area enclosed by a safety-hysteresis loop.

A larger HI indicates a stronger difference between how safety deteriorates and how it subsequently recovers.

Human Traffic Showed Stronger Effects

One of the clearest findings was that naturalistic traffic produced substantially higher and more variable safety hysteresis than controlled experiments.

Adaptive cruise control and autonomous vehicles were associated with lower safety HI, suggesting that automated systems may help reduce the build-up and persistence of risk during unstable traffic conditions.

The researchers did not simply compare automation with human driving. They also examined which traffic characteristics were most closely linked with stronger hysteresis.

At the vehicle-pair level, low speeds, large differences in speed between a following vehicle and the one ahead, and abrupt acceleration or braking were associated with higher HI.

At the platoon level, greater variation in traffic flow, density and average speed was linked with stronger safety hysteresis.

Braking Appears Especially Important

Simulation experiments helped the researchers explore possible mechanisms rather than relying only on correlations.

At the vehicle-pair level, stronger braking and lower speeds after deceleration increased safety HI, while acceleration rate and initial speed had much smaller independent effects.

At the platoon level, larger fluctuations in density, traffic flow and average speed consistently strengthened the hysteresis effect.

This suggests that smoothing sudden deceleration and reducing stop-and-go oscillations could be particularly important for improving safety.

Why The Findings Matter

Traditional traffic-safety indicators often capture risk at a particular instant. Safety hysteresis adds a time-dependent dimension, showing that the path into and out of a disturbance may also matter.

For road users, this helps explain why a traffic stream can remain unstable even after an obvious slowdown begins to clear. Drivers may still face elevated risk because vehicles are recovering at different rates and maintaining different gaps and speeds.

For traffic engineers, Safety Hysteresis Intensity could offer an additional system-level diagnostic tool. It may also help evaluate adaptive cruise control, autonomous-driving systems and traffic-management strategies designed to smooth braking and reduce oscillations.

The study does not suggest that automation removes road risk. Instead, its results indicate that more consistent vehicle responses may reduce one particular form of dynamic safety instability.

By focusing on how danger evolves over time rather than only how severe it appears at one moment, the research offers a more detailed way to understand why traffic disturbances can remain risky even after the road begins to move again.