High-Resolution Traffic Sensing with Probe Autonomous Vehicles: A Data-Driven Approach

Wei Ma, Wei Ma 0016, Sean Qian

TL;DR

This paper proposes a novel data-driven framework that leverages autonomous vehicles (AVs) as "probe sensors" to estimate high-resolution traffic states (flow, density, speed). By utilizing the rich sensor data from AVs, this approach aims to overcome the limitations of traditional traffic sensing methods and provide more accurate and ubiquitous traffic information, even with a low AV market penetration rate.

What Problem Does This Paper Address?

Traditional traffic sensing relies on fixed sensors (like loop detectors) or less capable moving sensors (like GPS-enabled taxis). Fixed sensors are expensive to install and maintain, leading to sparse coverage. Moving sensors, while offering better coverage, often only provide limited information (e.g., speed) and struggle with density estimation. The advent of autonomous vehicles (AVs) presents a unique opportunity: AVs are equipped with advanced sensors (LiDAR, cameras, radar) that not only perceive their own environment for safe navigation but can also be repurposed for traffic sensing. However, existing traffic state estimation (TSE) methods are not well-suited to the fragmented and complex data generated by AVs, especially during the transition period where AVs coexist with conventional vehicles. This paper addresses the challenge of effectively utilizing AV sensor data to achieve high-resolution traffic state estimation across different AV perception capabilities and market penetration rates.

Methods

The paper proposes a two-step, data-driven traffic sensing framework:

  1. Direct Observation: This step leverages the information directly observed by AVs based on their perception capabilities. The authors define three levels of perception (S1, S2, S3), ranging from tracking only the preceding vehicle (S1) to tracking all surrounding vehicles (S3). Each level allows for direct observation of traffic states (density and speed) in specific cells of a time-space region.
  2. Data-Driven Estimation: This step fills in the gaps where direct observation is not possible.
  3. Density Estimation: Matrix completion-based methods are used to estimate unobserved density values, treating the time-space density data as a matrix with missing entries.
  4. Speed Estimation: Both matrix completion-based and regression-based methods are employed. Regression models learn the relationship between density and speed, inspired by fundamental traffic flow diagrams.

The framework is designed to be flexible, accommodating different AV sensor configurations and market penetration rates.

Datasets

The proposed framework is validated using the Next Generation Simulation (NGSIM) data. This dataset provides high-resolution vehicle trajectory data collected using digital video cameras with a temporal resolution of 100ms. Experiments were conducted on three road networks: I-80, US-101, and Lankershim Boulevard.

Key Insights and Ideas

  • AVs as Ubiquitous Probe Sensors: The core idea is to harness the advanced sensing capabilities of AVs, which are already present for navigation, to provide a dense and continuous stream of traffic data. This transforms AVs into powerful "probe sensors."
  • Perception Levels Matter: The level of AV perception significantly impacts the amount of directly observable traffic data. Higher perception levels (S2, S3) allow for more comprehensive observation, reducing the reliance on estimation.
  • Data-Driven Approach is Key: Given the fragmented nature of AV data, a data-driven approach (matrix completion and regression) is more effective than model-driven methods for filling in missing traffic state information.
  • Density Estimation is More Challenging: Density estimation is more sensitive to missing data and sensor errors than speed estimation, as it requires more comprehensive vehicle detection.
  • Market Penetration is Crucial: While the framework shows promise even at low AV penetration rates (e.g., 5%), higher penetration rates significantly improve estimation accuracy, especially on complex roads.

Key Results

  • The proposed framework achieves accurate traffic state estimation on highways (I-80 and US-101) with a low AV market penetration rate (5%).
  • Estimation accuracy is generally higher for speed than for density.
  • For I-80 and US-101, perception level S2 is sufficient for accurate estimation, while Lankershim Boulevard requires S3.
  • The regression-based speed estimation model's coefficients align with traffic flow theory, showing that higher density generally leads to lower speed.
  • The framework demonstrates robustness to speed noise but is sensitive to detection missing rates and the frequency of distance measurement errors.
  • The proposed AV-based method outperforms traditional probe vehicle methods (conventional PVs and PVs with spacing measurement) in terms of estimation accuracy.

Limitations and Future Work

  • Privacy: The paper acknowledges that investigating the privacy implications of AVs sharing observed information with a data center is an interesting future research direction.
  • Network-Level Estimation: The current framework focuses on road-level traffic state estimation. Extending it to network-wide estimation is a key area for future work.
  • Traffic Simulation Environment: The authors plan to develop a traffic simulation environment to enable comprehensive analysis under various traffic conditions.
  • Data Granularity: While the framework performs well with larger discretization sizes, it highlights the challenge of achieving high-resolution estimation when data is sparse.
  • Complex Scenarios: The framework underperforms on the complex arterial road (Lankershim Boulevard), suggesting that more sophisticated methods might be needed for highly complex urban environments.

Summary updated

Abstract

Recent decades have witnessed the breakthrough of autonomous vehicles (AVs), and the sensing capabilities of AVs have been dramatically improved. Various sensors installed on AVs will be collecting massive data and perceiving the surrounding traffic continuously. In fact, a fleet of AVs can serve as floating (or probe) sensors, which can be utilized to infer traffic information while cruising around the roadway networks. Unlike conventional traffic sensing methods relying on fixed location sensors or moving sensors that acquire only the information of their carrying vehicle, this paper leverages data from AVs carrying sensors for not only the information of the AVs, but also the characteristics of the surrounding traffic. A high-resolution data-driven traffic sensing framework is proposed, which estimates the fundamental traffic state characteristics, namely, flow, density and speed in high spatio-temporal resolutions and of each lane on a general road, and it is developed under different levels of AV perception capabilities and for any AV market penetration rate. Experimental results show that the proposed method achieves high accuracy even with a low AV market penetration rate. This study would help policymakers and private sectors (e.g., Waymo) to understand the values of massive data collected by AVs in traffic operation and management.
BibTeX
@article{ma2021high,
  author = {Wei Ma and Wei Ma and Sean Qian},
  title = {{High-Resolution Traffic Sensing with Probe Autonomous Vehicles: A Data-Driven Approach}},
  journal = {Sensors},
  year = {2021},
  doi = {10.3390/s21020464},
}
Figure