Source-linked AI summary
On some experimental features of car-following behavior and how to model them
Rui Jiang, Mao-Bin Hu, H. M. Zhang, Zi-You Gao, Bin Jia, Qing-Song Wu
TL;DR
The paper addresses whether car-following traffic states obey a unique speed–spacing relation, using high-precision platoon experiments and model simulations. The experiments show two-dimensional speed–spacing states and differing platoon lengths at similar average velocities, while a driver-insensitivity model reproduces the experimental results well.
Problem
Traffic-flow theory debates the speed–spacing relation, but available traffic data lack continuous, high-fidelity observations of individual-vehicle dynamics.
Method
The authors extend 25-car-platoon experiments with high-precision GPS measurements and test established and newly proposed car-following models.
Results
Traffic states span a 2D speed–spacing region, and platoon length can differ significantly at essentially the same average velocity; the new model reproduces the experiments well.
Takeaways & Limitations
A single speed–spacing curve is inadequate for representing the observed variability and imprecision of human driving.
Takeaways & Limitations
The model does not reproduce some experimental behavior because peculiar driver behavior and vehicle-driver heterogeneity are not fully represented.
Abstract
from arXiv · showhide
We have carried out car-following experiments with a 25-car-platoon on an open road section to study the relation between a car's speed and its spacing under various traffic conditions, in the hope to resolve a controversy surrounding this fundamental relation of vehicular traffic. In this paper we extend our previous analysis of these experiments, and report new experimental findings. In particular, we reveal that the platoon length (hence the average spacing within a platoon) might be significantly different even if the average velocity of the platoon is essentially the same. The findings further demonstrate that the traffic states span a 2D region in the speed-spacing (or density) plane. The common practice of using a single speed-spacing curve to model vehicular traffic ignores the variability and imprecision of human driving and is therefore inadequate. We have proposed a car-following model based on a mechanism that in certain ranges of speed and spacing, drivers are insensitive to the changes in spacing when the velocity differences between cars are small. It was shown that the model can reproduce the experimental results well.
1. Introduction
The introduction frames unresolved debates in traffic-flow theory around the speed–spacing relation and the need for high-fidelity vehicle-trajectory data. Prior models commonly assume a unique steady-state speed–spacing relation, while the authors’ experiments and proposed models challenge and extend that assumption.
- Motivation: Traffic-flow theory remains unsettled, including debate over whether congested traffic requires a two-dimensional flow-density description.Available data and competing theories have not resolved the controversy.
- Prior models: Many established car-following models assume a unique steady-state relationship V(Δx) between vehicle speed and spacing.This assumption appears explicitly or implicitly in Newell, intelligent driver, full velocity difference, and related models.
- Motivation: Sparse point-sensor data and short, site-specific camera observations cannot capture complete disturbance evolution across traffic congestion.The paper motivates precise spatiotemporal trajectories of individual vehicles, especially for tracking disturbance genesis and growth.
- Experiments: A 25-car open-road experiment recorded individual vehicle locations and velocities with high-precision GPS while varying the lead vehicle’s speed.The design enabled study of disturbance evolution under multiple traffic situations.
- Experimental findings: Spacing can change substantially while adjacent vehicles maintain nearly constant speeds and very small velocity differences, contradicting a unique speed–spacing relation.The authors propose driver insensitivity within certain spacing ranges and time-varying preferred spacing as possible mechanisms.
- Modeling contribution: Models based on variable preferred spacing reproduce observed disturbance-growth patterns more consistently than the OV, FVD, and ID two-phase models.The paper additionally proposes a model based on driver insensitivity to spacing changes when velocity differences are small.
2. Experimental setup and results
Experiments used GPS-equipped car platoons to examine how speed and spacing vary across traffic conditions. The results show substantial spacing variability, including different average spacings and platoon lengths at nearly identical average velocities.
- Experimental setup: A 25-car platoon was measured on a 3.2 km road section using high-precision GPS devices recording each car’s location and velocity every 0.1 second.Location errors were within ±1 m and velocity errors within ±1 km/h; drivers followed without overtaking.
- Previous experimental results: Spacing fluctuated significantly even when a following car and its predecessor maintained essentially constant speeds with very small velocity differences.This behavior was observed in the earlier 25-car experiments and challenges a unique speed-spacing relationship.
- Previous experimental results: At low traffic-flow speeds, cars in the rear of the 25-car platoon developed stop-and-go motion, while velocity standard deviation increased along the platoon in a concave or linear pattern.The speed-variation curve would bend downward in a much longer platoon because physical speed limits cap tail fluctuations.
- New experimental results: Spacing ranged from 30 m to 60 m near 51 km/h but fluctuated below 25 m near 56 km/h in two runs involving the second car.The higher-speed case had a smaller minimum spacing than the lower-speed case.
- New experimental results: At about 46 km/h average velocity, two runs showed substantially different average spacings for the 15th car in the platoon.The velocity fluctuated between about 35 km/h and 58 km/h in both runs.
- New experimental results: Spacing fluctuations were much larger than velocity fluctuations in the high-speed 3-car-platoon experiments.Across four panels, σv/vave ranged from 0.012 to 0.048, whereas σ∆x/∆xave ranged from 0.139 to 0.242.
- New experimental results: Different runs could produce significantly different platoon lengths and average spacings even when the platoon’s average velocity was approximately the same.Platoon length relates to average spacing through s = L/(N-1), with density ρ = 1/s for N=25.
3. Simulation results of GM models and Gipps model
Simulations show that GM and Gipps car-following models fail to reproduce key experimental features, particularly the observed concave or linear growth of velocity variation along the platoon.
- Simulation setup: The simulations test whether GM and Gipps models reproduce experimentally observed traffic-flow features.The comparison focuses on the platoon’s velocity-variation profile and stability across leading-car speeds.
- GM models: GM model (1) produces a velocity-variation curve that initially increases convexly, unlike the experimental results.At low leading-car speeds, speed bounds can make the rear portion concave, especially in longer platoons.
- GM models: GM model (2) is stable at some leading-car speeds but shows convex-then-concave velocity-variation growth at 7 and 15 km/h.For the parameter set with m = 1, stopped cars require a model modification because they otherwise never restart.
- GM models: At leading-car speeds of 30 and 40 km/h, one GM simulation becomes very unstable, with unrealistically strong fluctuations in the second car’s velocity.The reported parameters for this case include λ = 0.5 s−1.
- Conclusion: Overall, the GM models and the Gipps model do not reproduce the observed experimental features of traffic flow.The mismatch concerns both instability behavior and the shape of velocity-variation growth along the platoon.
- Gipps model: The Gipps model is stable at 7 and 15 km/h but unstable at 30, 40, and 50 km/h, where σv increases convexly.Its simulated behavior is therefore inconsistent with the experimental results.
4. A new model and simulation results
The paper proposes a car-following model with a two-dimensional velocity-spacing region where drivers may ignore small velocity differences and spacing changes. Simulations reproduce key experimental patterns, including differing spacing and platoon length at similar average velocities, though some discrepancies remain.
- Model mechanism: The model assumes drivers are insensitive to spacing changes within certain velocity-spacing ranges when velocity differences are small.When the velocity difference is below a threshold, acceleration changes randomly because drivers cannot maintain exactly constant velocity.
- Model mechanism: Outside the two-dimensional region R, drivers respond to both velocity difference and spacing through the FVD model.The region is bounded by five straight lines, and the OV function defines the spacing-dependent optimal velocity.
- Relation to prior models: The model extends earlier two-dimensional traffic-state ideas while using an insensitivity mechanism for small velocity differences.The paper contrasts this approach with models that assume a unique speed-spacing relationship.
- Simulation results: The simulated velocity standard deviations, spatiotemporal stripes, and speed distributions agree qualitatively and quantitatively with experimental results.The model uses a 25-car platoon and reproduces the observed velocity patterns; agreement is weaker at 7 km/h.
- Simulation results: Simulations reproduce large spacing fluctuations despite small velocity fluctuations, including different average spacings across runs with nearly identical average velocities.These patterns are reported for individual cars and are consistent with the experiments.
- Simulation results: The simulated platoon length can differ across runs even when the leading-car movement and average platoon velocity are essentially the same.The difference is smaller than in the experiments and becomes small at lower leading-car speeds because the two-dimensional region narrows.
5. Conclusion
The conclusion reports experimental evidence that speed and spacing do not follow a unique relationship and that platoon length can vary at similar average speeds. It proposes an insensitivity-based model that reproduces the experiments well, while calling for larger experiments to test traffic evolution and the models further.
- Experimental findings: Experiments found that spacing can be smaller at larger speeds and that average spacing and spacing fluctuations can differ at nearly identical average speeds.These findings come from further analysis of 25-car-platoon experiments.
- Experimental findings: Platoon length might differ significantly even when the platoon's average velocity is essentially the same.A separate 3-car high-speed experiment showed similar car-following results to the 25-car experiment.
- Model conclusion: The proposed model assumes drivers are insensitive to spacing changes when velocity differences are small and reproduces the experimental results well.Drivers adjust spacing only when it becomes sufficiently large or small to require acceleration or deceleration.
- Future work: Larger experiments with longer road sections and larger platoons are needed to study traffic evolution and examine the proposed models.The authors report additional 50-car and 11-car experiments for future publications.