Data-driven and Empirical Research for Emerging Mixed Traffic of Automated Vehicles and Human-driven Vehicles

Title: From Micro to Macro: Identifying automated vehicles and inferring their fundamental diagram


The heterogeneity of drivers plays a pivotal role in the intricate dynamics observed in traffic behavior. Experimental evidence underscores the distinct driving patterns exhibited by humans (HD) and automated controllers (AD). A robust deep-learning framework estimates the driver type from mere seconds of historical raw trajectory data. 

Furthermore, we study the Fundamental Diagram (FD) of heterogeneous vehicular platoons. Strong aggregation of observations yields minimal scatter, it demonstrates a remarkably consistent relation between the traffic variables and a clear triangular shape for autonomously-driven vehicles.

The penetration rate of Adaptive Cruise Control (ACC) systems on commercial vehicles is constantly increasing. ACC systems attract a lot of research interest. They are considered a proxy to future autonomous longitudinal vehicle movement, which is expected to bring significant changes in future transport networks in sustainability, traffic flow, safety, and other dimensions. JRC conducted several experimental car-following campaigns to understand how commercial ACC systems operate and their impact on motorway traffic. This talk is about the behavior of car-platoons equipped with ACC, the similarities and differences with human drivers and the essential role of connectivity.

Current Research

Research and development in traffic engineering with a focus on intelligent transportation systems, partially and fully automated vehicles, traffic flow, energy consumption and sustainability. Research activities involve modelling and simulation, observation-based analysis, analytical studies and machine learning applications. 

Traffic flow, modeling and simulation

Traffic Flow: Understanding the impact of heterogeneity in longitudinal driving behaviors on freeway networks.

Traffic Estimation, Prediction: Quantifying the benefits of data fusion between models and observations for better traffic state understanding and prediction.

Traffic Control: Investigating the use of using Connected and Automated Vehicles as mobile actuators for traffic estimation and control.

Traffic Safety: Understanding the different safety perceptions between automated driver such as Adaptive Cruise Control and Humans.

Energy demand and CO2 emissions: Quantifying and comparing the energy needs of (partially or fully) automated systems and human drivers.

Modeling vehicle dynamics: Proposing modeling approaches for incorporating different powertrain behaviors in microsimulation leading to more precise reproduction of electric, hybrid and vehicles with Internal Combustion Engines dynamics.

Understanding reaction time differences between drivers: Response Time and Time Headway of commercially-available Adaptive Cruise Control (ACC) Systems.

String Stability of automated vehicles: Investigating string instability issues of automated vehicles, compare them with human driving behaviors and project such impacts on traffic flow.

Energy Demand of ACC Systems: Quantifying the energy consumption of automated vehicles through empirical observations

Machine learning applications in traffic engineering: Exploring solutions for error pattern learning between models and observations through machine learning. Identifying driver behaviors from empirical data designing ML frameworks with high generalization value.

Past Research

Data Mining

Infer of the itinerary of cargo transported in shipping containers based on a large, heterogeneous and noisy dataset of Container Status Messages. Such itinerary information can be used to improve the risk analysis performed by authorities in their effort to secure the global trade and produce reliable anti-fraud risk indicators.

Quality Assessment of Risk Indicators

We proposed a quality assessment framework that combines quantitative and qualitative domain specific metrics to evaluate the quality of large datasets of Container Trip Information records and to provide a more complete feedback on which aspects need to be revised to improve the quality of the output data.

Automatic archaeological sherd classification

Sherds that are found in the field usually have little to no visible textual information such as symbols, graphs, or marks on them. This makes manual classification an extremely difficult and time-consuming task for conservators and archaeologists. For a bunch of sherds found in the field, an expert identifies different classes and indicates at least one representative sherd for each class (training sample). We developed an image processing technique that uses local features based on color and texture information to automatically cluster similar pottery fragments and speed up the archaeologists' labelling tasks.

Jigsaw puzzle solving

This work was on providing an automatic jigsaw puzzle solution without any initial restriction about the shape of pieces, the number of neighbor pieces, etc. The proposed technique uses both curve- and color-matching similarity features and applies a recurrent algorithm that matches puzzle pieces in pairs aiming to restore the original image.

Analysis, Segmentation, Restoration of degraded document images

We worked on global and local skew detection in complex color documents with both contextual and graphical information, segmentation of historical machine-printed documents for the detection of characters, words and sentences, as well as document binarization and color reduction.