[Paper Review] Driver Digital Twin for Online Prediction of Personalized Lane Change Behavior

Xishun Liao, Xuanpeng Zhao+6 more|arXiv (Cornell University)|Nov 2, 2022

Vehicle emissions and performance4 citations

TL;DR

This paper proposes a Driver Digital Twin (DDT) system that enables connected and automated vehicles (CAVs) to predict personalized lane change behavior in real time using a vehicle-edge-cloud architecture. By modeling individual drivers via historical data on the cloud and processing real-time data on the edge, the system achieves 94.3% accuracy in detecting lane change intentions 6 seconds before execution, with a mean trajectory error of 1.03 meters over a 4-second window, outperforming general models by 27.8% in accuracy.

ABSTRACT

Connected and automated vehicles (CAVs) are supposed to share the road with human-driven vehicles (HDVs) in a foreseeable future. Therefore, considering the mixed traffic environment is more pragmatic, as the well-planned operation of CAVs may be interrupted by HDVs. In the circumstance that human behaviors have significant impacts, CAVs need to understand HDV behaviors to make safe actions. In this study, we develop a Driver Digital Twin (DDT) for the online prediction of personalized lane change behavior, allowing CAVs to predict surrounding vehicles' behaviors with the help of the digital twin technology. DDT is deployed on a vehicle-edge-cloud architecture, where the cloud server models the driver behavior for each HDV based on the historical naturalistic driving data, while the edge server processes the real-time data from each driver with his/her digital twin on the cloud to predict the lane change maneuver. The proposed system is first evaluated on a human-in-the-loop co-simulation platform, and then in a field implementation with three passenger vehicles connected through the 4G/LTE cellular network. The lane change intention can be recognized in 6 seconds on average before the vehicle crosses the lane separation line, and the Mean Euclidean Distance between the predicted trajectory and GPS ground truth is 1.03 meters within a 4-second prediction window. Compared to the general model, using a personalized model can improve prediction accuracy by 27.8%. The demonstration video of the proposed system can be watched at https://youtu.be/5cbsabgIOdM.

Motivation & Objective

• To address the challenge of safe and efficient operation of connected and automated vehicles (CAVs) in mixed traffic with human-driven vehicles (HDVs).
• To develop a personalized, real-time prediction system for lane change behavior of HDVs to improve CAV decision-making.
• To enable online, scalable, and adaptive driver behavior modeling through a digital twin framework deployed across vehicle-edge-cloud infrastructure.
• To validate the system’s performance in both simulated and real-world field environments under real-time network conditions.

Proposed method

• A cloud-based digital twin is created for each HDV using historical naturalistic driving data to model individual driver behavior patterns.
• Real-time vehicle data from each driver is processed on the edge server using their corresponding digital twin model to predict lane change maneuvers.
• The system employs a hierarchical architecture: cloud for long-term behavior modeling, edge for real-time inference, and vehicle for data acquisition.
• A personalized prediction model is trained on individual driver trajectories, improving accuracy over generic models.
• The system uses a 4-second prediction window and evaluates performance using Mean Euclidean Distance (MED) between predicted and GPS-verified trajectories.
• The architecture is deployed over a 4G/LTE network to ensure real-time communication in field trials.

Experimental results

Research questions

• RQ1 Can a digital twin framework effectively predict individual driver lane change intentions in real time within a mixed-traffic environment?
• RQ2 How does personalized modeling via digital twins compare to generic models in predicting lane change behavior?
• RQ3 What is the temporal accuracy of lane change intention prediction using the proposed edge-cloud architecture?
• RQ4 How does the system perform in real-world field conditions with actual human drivers and network latency?
• RQ5 What is the trajectory prediction accuracy of the system compared to ground truth GPS data?

Key findings

• The system detects lane change intentions with 94.3% accuracy on average 6 seconds before the vehicle crosses the lane separation line.
• The Mean Euclidean Distance (MED) between predicted and ground-truth trajectories is 1.03 meters over a 4-second prediction window.
• Using personalized models improves prediction accuracy by 27.8% compared to general (non-personalized) models.
• The system achieves real-time performance with a mean prediction latency of 6 seconds before lane change initiation.
• Field tests with three passenger vehicles over a 4G/LTE network confirm the system’s robustness and scalability in real-world conditions.
• The human-in-the-loop co-simulation platform successfully validates the system’s behavior prediction under diverse driving scenarios.