Platoon Fundamental Diagram estimation can be Markovian: evidence from human- and self-driven vehicle trajectories

Michail A. Makridis, Anastasios Kouvelas, Jorge A. LavalarXiv 2024 · arXiv

TL;DR

Researchers have developed a simple, "Markovian" method to estimate traffic Fundamental Diagrams (FDs) using only two consecutive snapshots of vehicle platoon trajectories. Their findings reveal that while Adaptive Cruise Control (ACC) can nearly double road capacity compared to human drivers, it introduces potential safety risks through aggressive "backward wave speeds" that demand further investigation.

What Problem Does This Paper Address?

The Fundamental Diagram (FD)—the mathematical relationship between traffic flow, density, and speed—is the cornerstone of transportation engineering. However, deriving a clean FD from real-world trajectory data is notoriously difficult. Traditional methods often require "stationary" traffic conditions (where speeds aren't changing) or require researchers to guess the "backward wave speed" (how fast a slowdown ripples back through a line of cars).

Furthermore, as we transition to mixed traffic, we need to understand how different "driver modes"—human, Adaptive Cruise Control (ACC), and Cooperative ACC (CACC)—change these diagrams. Existing literature lacks a robust, generalized way to compare these modes across varying platoon lengths, vehicle types, and data qualities.

Methods

The authors propose a remarkably lean approach. Instead of looking for long periods of steady-state traffic, they use a Markovian approach: the traffic state at any moment is estimated using only two consecutive observations (typically 0.1 seconds apart).

Key technical steps include:

  1. Dynamic Trapezoidal Areas: They calculate flow, density, and speed using Edie’s generalized definitions within a time-dependent trapezoidal space-time area defined by the platoon's front and rear vehicles.
  2. Bivariate Data Aggregation: To cut through the massive "scatter" (noise) in raw trajectory data, they employ a binning technique. They divide the density range into small intervals and average all observed flow values within each bin to find the "fundamental" state.
Figure 1: Schematic representation of an observed platoon and a dynamic area for instantaneous traffic state estimation.

This method is "Markovian" because the estimated state depends only on the current transition between two time steps, making it computationally efficient and suitable for real-time monitoring.

Datasets

The study draws from eleven different experimental campaigns across three major sources:

  • OpenACC Dataset: Includes human-driven and ACC-driven experiments from test tracks like AstaZero and ZalaZone.
  • CARMA Dataset: Provides rare empirical data on Cooperative ACC (CACC) platoons using V2V (vehicle-to-vehicle) communication.
  • Vanderbilt University Dataset: Features multi-vehicle ACC platoon tests.

These datasets cover a wide range of conditions, from two-vehicle pairs to platoons of eleven, and various ACC headway settings (minimum, median, and maximum).

Key Insights and Ideas

  • The Triangular Truth of Automation: For ACC-driven vehicles, the FD follows a remarkably consistent triangular shape. This suggests that automated systems are much more predictable and "mathematically well-behaved" than humans.
  • Human "Agility" vs. Automation: Human drivers don't follow a simple triangle. They are more cautious (larger gaps) at high speeds but become more "agile" and take higher risks (shorter gaps) in congested, low-speed traffic. This results in a smoother, curved FD compared to the sharp peak of ACC.
  • The Connectivity Boost: Preliminary data on CACC (connected vehicles) shows a massive potential for capacity. By sharing information, these vehicles can maintain high speeds even at very high densities, pushing flow rates toward 4000 vehicles per hour.
  • The Safety Trade-off: While setting ACC to "minimum headway" increases capacity by roughly 90% over the "maximum" setting, it results in very high backward wave speeds (over 30 km/h). This suggests that a small disturbance at the front of an ACC platoon could trigger a violent, potentially unsafe reaction at the back.

Key Results

The proposed method was validated against state-of-the-art techniques, showing high resemblance but with significantly less computational complexity.

Figure 3: Flow-Density plot with the proposed method and the measurement method proposed by Laval (2011) after the application of the Shear transformation described in Maiti et al. (2023) and the aggregation process described in this paper.
  • Capacity Gains: ACC with minimum distance settings achieves nearly double the capacity of maximum settings.
  • Robustness: The resulting FDs proved to be invariant to platoon length and vehicle characteristics, meaning the method captures the inherent "DNA" of the driver mode rather than the specifics of the test.
  • Platoon States: By analyzing acceleration and deceleration separately, the authors found that "stable" cruising states actually sit right in the middle of the FD, and you need the transitions (speeding up/slowing down) to map the full diagram.

Limitations and Future Work

While the results are promising, the authors note a few hurdles:

  • Data Scarcity: There is still very little empirical data for CACC and "medium" ACC settings, making those specific FDs less certain.
  • Idealized Environments: Most data comes from test tracks. Real-world factors like lane changes, grades, and weather were not the primary focus.
  • Safety Deep-Dive: The high backward wave speeds in ACC platoons are a red flag. The authors call for future research to investigate whether these high-capacity settings are actually stable enough for public roads.

This work provides a powerful new tool for engineers to monitor mixed-traffic streams in real-time, paving the way for more intelligent traffic management as self-driving features become standard.

Summary updated

Abstract

We propose a simple and effective method to derive the Fundamental Diagram (FD) from platoon vehicle trajectories. Average traffic state variables are computed using Edie's generalized definitions within time-dependent trapezoidal space-time areas. To obtain a clear FD, we employ a bivariate data aggregation technique to eliminate scatter. Our findings are as follows: (i) The proposed method demonstrates a remarkably consistent relation between the traffic variables and a clear triangular shape for autonomously-driven vehicles. (ii) The FDs are invariant to several factors of heterogeneity such as the platoon length, vehicle characteristics, road particularities, and data acquisition accuracy. (iii) ACC-driven vehicle platoons with minimum headway setting achieve much higher capacity, roughly 90\% than those with a large headway setting. (iv) Connectivity might increase capacity. (v) Human drivers have a wider near-capacity operation area, showing different behaviors at high speeds than low ones, and (vi) Safety concerns might arise due to high values of backward wave speed for ACC-driven vehicles. Comparative analysis with the state-of-the-art confirms the validity of our approach. The proposed method stands out due to its simplicity and accuracy, which paves the way for practical applications in real-time traffic flow monitoring and control within modern intelligent transportation systems.
BibTeX
@misc{makridis2024platoon,
  author = {Michail A. Makridis and Anastasios Kouvelas and Jorge A. Laval},
  title = {{Platoon Fundamental Diagram estimation can be Markovian: evidence from human- and self-driven vehicle trajectories}},
  year = {2024},
  eprint = {2401.17065},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2401.17065},
}
Figure