Traffic and granular flow: the role of data and technology in the understanding of particle dynamics

Introduction

From individual Connected and Automated vehicles (CAVs) to system-level Smart City applications, technology is playing a more dominant role in research associated with traffic and granular flow modeling. With advanced detection instrumentation, the positioning and the movement of objects (ranging from nanoparticles to transport vehicles) and humans (pedestrians, drivers/travelers, etc.) are accurately characterized and monitored. For example, (1) the properties of objects are modified to offer different “smart” functionalities for the benefit of societies. As an application, materials are being studied and engineered for energy harvesting and filtering applications. Conversely, (2) changing the interactions between humans/machines can produce more efficient system-level behavior. As an application, different generations of CAVs with multiple levels of automation are being deployed along with human-driven vehicles (HDVs) to reach more efficient and safer transportation systems.

Focusing on the transportation sector that involves both objects and humans, concepts of Intelligent Transportation Systems (ITS) are being further extended and utilized to make rural and urban surface transportation networks seamlessly coordinated, sustainable, equitable, and safe. Mobility is not simply related to moving people and goods between an origin and destination. Instead, transportation systems are perceived as a component of a system of interdependent systems and new technologies are needed to tackle challenges related to human-machine interactions, big data analytics, traffic inter-modality and heterogeneity, ethical and equitable artificial intelligence (AI), privacy and security in human subject detection, etc. As an example, in the United States, the total number of micro-mobility trips has increased from 0.321 million trips in 2010 to 84 million trips in 2018 and 136 million trips in 2019 (NACTO, 2019). These trips are leading to more heterogeneity in urban traffic with more pedestrians, scooters, and cyclists occupying city roadway networks. Measuring the safety and the mobility performance of such traffic is a challenging topic that requires new tools including video detection algorithms. Such tools should supplement existing traffic detector and probe vehicle measurements traditionally applied to capture the microscopic (i.e., vehicular trajectories) and the macroscopic (roadway flow, density, and speed) traffic behavior.

This special issue brings together recent theoretical developments and empirical investigations on heterogeneous traffic flow that is composed of cars, pedestrians, trucks, motorcycles, etc. Special focus is on data-driven empirical approaches (i.e., based on floating car data, probe vehicle data, and video extracted data) and latest modeling analytical approaches (ranging from improvements on traditional microscopic traffic modeling to adapting AI models for traffic flow prediction). The seven papers published in this “Traffic and Granular Flow − 2017” special issue are selected from around 20 papers that were considered and reviewed. These 20 papers consist a portion of the 45 papers that were initially invited to be submitted to the Journal of Intelligent Transportation Systems (JITS) given the relevance of the research areas tackled. The 45 papers were among 90 submissions (57 lectern session papers and 36 poster session papers) presented at the 12th Traffic and Granular (TGF) Conference that was held at the George Washington University, Washington, DC, USA, between the 20th of July and the 22nd of July, 2017. The theme of the 12th TGF conference was: “From Molecular Interactions to Internet of Things and Smart Cities: The Role of Technology in the Understanding and the Evolution of Particle Dynamics.”

The seven final published papers are associated with vehicular traffic (i.e., motorized vehicles) that is lane-based with a focus on empirical characterization (Dülgar et al., 2019), data-driven microscopic control (Molzahn et al., 2019), and simulation (Wegerle et al., 2019). Two papers deal with heterogeneous traffic modeling that involves different types of vehicles: the first paper focuses on data extraction and analysis for calibration purposes (Raju et al., 2019) and the second paper formulates a traffic flow model for non-lane-based heterogeneous traffic (Gaddam & Rao, 2020). The final two papers deal with pedestrian traffic: the first paper focuses on predicting the movement of pedestrians using an AI model (Tordeux et al., 2019) while the second paper applies an automated video detection algorithm to capture high-density pedestrian flow dynamics (Baqui et al., 2019).

We hope that the selected papers can advance the understanding of heterogeneous traffic dynamics (i) by adopting improved modeling techniques and (ii) by extracting more accurate data and analyzing the corresponding real-world congested and non-congested phenomena. Collectively, they could help address existing and future challenges around the use of technologies to reach a more efficient multimodal surface transportation system.

Recent advances in traffic flow modeling and detection

In Dülgar et al. (2019), the authors were motivated by the lack of empirical characterization of traffic “break-down” as a two-way transition between different traffic states. The authors adopt the classification of Kerner (2015) that divides the traffic fundamental diagram into three regions. The traffic breakdown is then seen as a transition between a Free-flow Region (designated by F) and a Synchronized Region (designated by S): the “nucleation” in the F-S transition leads to a traffic breakdown that is governed in its turn by the S-F “instability.” The authors utilize large amounts of probe vehicle data obtained from a fleet of vehicles (more than two-million vehicles) deployed in Europe. The data are in the form of “pearl chains” with global positioning system (GPS) data transformed into trajectories and speeds with a 5/10 s time resolution and a 2 m/s speed inaccuracy threshold. A traffic phase transition algorithm is built by the research team relying on identifying different speed-time transition points. Looking at different F-S-F transitions observed in 67 days from 2015 until 2018, the authors identified a Moving Synchronized Flow Pattern (MSP) that builds and dissolves (i.e., “dissolving MSP”) with time. The following conclusions were found: (i) the sequences of F-S-F transitions differ at the same bottleneck location; (ii) the time of occurrence of a F-S-F transition is random; (iii) the duration between the F-S and the S-F transitions and the S-F and the F-S transitions are random. The randomness in the aforementioned transitions support data-driven traffic flow control where different measures can be applied at bottlenecks when entering a synchronized traffic region.

Molzahn et al. (2019) further expanded on the aforementioned empirical characterization by developing a jam warning system to avoid deceleration behavior that may lead to rear-end collisions. Utilizing also probe vehicle GPS data, the concept of “jam wall” is introduced: a microscopic speed drop experienced when entering a synchronized flow regime or a jam flow regime. The GPS data feeds into a trajectory reconstruction algorithm and a time-space function of the speed drop is developed. This function is bounded in a time-space diagram by traffic phase transition points from one end and by minimum bottleneck speed points on the other hand. Focusing on on-ramp bottlenecks in Germany, the authors consistently show the presence of the “wall” (i.e., the “generality of the concept”) despite the fact that the form (i.e., the “thickness”) of such “wall” may be a complex function to formulate. The “jam wall” changes with the traffic states encountered downstream: the danger associated with the corresponding transition (i.e., when encountering a wide moving jam phase vs. when encountering a slower synchronized flow phase) dictates the need of the proposed data-based traffic control approach. Alternative shorter-term prediction methods (possibly AI methods) may be needed to better capture the jam fronts when encountering downstream synchronized flow. Higher-resolution data with larger probe vehicle market penetrations are required for answering such need.

Wegerle et al. (2019) tried to address and investigate the shortcomings of high-resolution data availability in the aforementioned research (Molzahn et al., 2019) by adopting simulation to create synthetic data and to better predict the moving bottleneck (MB) formation. MBs are harder to identify and locate if compared to roadway bottlenecks as they can be created unexpectedly and may cause problems for non-connected autonomous vehicles (i.e., if no proper vehicle to vehicle [V2V] and vehicle to infrastructure [V2I] communication is established). The microscopic Kerner–Klenov stochastic traffic flow model is used to reproduce vehicular trajectory data as it can reproduce the “nucleation” process mentioned in Dülgar et al. (2019). The authors were able to predict the MBs’ formation and a statistical analysis on the quality of such predictions was presented: MBs can be predicted if two traffic phase transition points are found with a 1% to 2% of probe vehicle market penetration rate. Some future considerations such as communication latency and GPS data accuracy may need to be addressed to translate the proposed control method into reality.

The previous three papers focus on data-driven approaches to capture different homogeneous traffic phenomena and to offer connected control measures to alleviate congestion and to improve safety. Raju et al. (2019) attempted instead to analyze and model heterogeneous following behaviors with the presence of two-wheel/three-wheel vehicles, cars, buses, trucks, and light-commercial vehicles (LCVs) sharing the roadway (i.e., a mixed environment). The authors utilize trajectory data extracted from videos monitoring Indian roadway sections (a Delhi-Gurgaon road segment and a Chennai urban arterial). The key objective is to identify different types of leader-follower pairs and to test the feasibility of using existing car-following models to capture the resulting interactions; three car-following models are calibrated using a Genetic Algorithm approach: the Gipps Model (Gipps, 1981), the Bando Model (Bando et al., 1998), the Intelligent Driver Model (IDM) (Treiber et al., 2000), and the Wiedemann Model (Wiedemann, 1974). The Gipps model, the Wiedemann Model (mainly the Wiedemann 99 model), and the IDM model show acceptable results in replicating the following behavior in heterogeneous traffic with hysteresis formation. The main issue remains associated with data availability to perform such calibration and the number of scenarios needed to produce different parametric values accounting for different leader-follower pairs.

Gaddam and Rao (2020) adopted a more analytical approach developing a new continuum model for non-lane-based heterogeneous traffic considering both the lateral and the longitudinal movement of vehicles. A “density dependent propagation speed, a viscosity term, and a frictional clearance term” are integrated, and the resulting model is analyzed for feasibility. Special focus is dedicated to linear stability analysis and congestion propagation and dissipation. The simulation results show a capacity increase, an improved stability, and faster traffic perturbation dissipation when allowing for non-lane-based interactions (lateral and longitudinal simultaneous interactions) with heterogeneous traffic. These results are consistent with empirical observations and may motivate adopting the proposed model to manage traffic in congested urban settings and to design alternative transport policy/solutions in rising economies with serious surface roadway traffic problems.

The final two papers in this special issue address topics associated with pedestrian traffic. Tordeux et al. (2019) used a speed-based Wiedemann model and four different types (differing in the number of neurons and the hidden layers) of artificial neural network (NN) models to predict the speed of pedestrians in two types of geometry: a corridor geometry where pedestrians are moving in loops and a bottleneck geometry where pedestrians have to exist a given area. With proper training, it is shown that the NN models can differentiate between different traffic scenarios and improve on the speed prediction (i.e., lower speed mean square errors) if compared to the traditional speed-based traffic models.

Baqui et al. (2019) offered an automated Closed Circuit Television (CCTV) image processing system to extract pedestrian crowd data and to predict the corresponding flow dynamics in dense traffic scenarios. The authors integrated an image feature extraction module, a machine learning module, a count module, a coordinate transformation module for data collection. The data were then used alongside a pedestrian flow model (PEDFLOW) (Lohner, 2010) for prediction purposes. The authors chose imagery from the Mecca pilgrimage (Saudi Arabia) to demonstrate the feasibility and the potential of the proposed framework. The maximum error in velocity estimation is 0.2 m/s with a 95% confidence interval and the upper bound of the 95% count error is 20%. The predicted and the actual speed/density data adhere to the trends expected to be seen in a congested pedestrian traffic fundamental diagram.

Summary and further research needs

Tremendous efforts have been made (1) to collect data and (2) to deploy traditional traffic modeling and AI tools with the objective of creating more efficient heterogeneous surface transportation networks. In this special issue, (i) probe vehicle GPS data, (ii) synthetic simulation data, (iii) CCTV camera data, and (iv) video data have been utilized to analyze different particle interactions/dynamics and to control different types of vehicles/pedestrians in multiple settings. On the other hand, (i) traditional and newly developed traffic models (such as the Wiedemann model, the IDM Model, the PEDFLOW Model, the Kerner–Klenov stochastic traffic flow model, etc.) and (ii) artificial intelligent tools (such as NNs and machine learning regressors) have been adopted to offer additional understanding for different traffic phenomena. In line with the theme of this special issue (Traffic and Granular Flow: The Role of Data and Technology in the Understanding of Particle Dynamics), we hope that the papers published can provide theoretical and practical tools to bridge the gap between the technologies being introduced in our transportation systems and the research conducted by the traffic flow community on connected multimodal roadway environments.

Despite the contributions offered in this special issue, significant additional efforts are needed. More exploration of traffic interactions in shared right of way (involving pedestrians, cyclists, scooters, and vehicles) need to be performed. Additional novel technologies (LIDARs, Bluetooth sensors, cell-phone accelerators, etc.) may be researched and applied to extract trajectories and to develop real-time decentralized traffic control and warning strategies. More consideration of the human component is needed as the interactions between the travelers and the offered technologies are not trivial: collecting unobserved data related to perception and judgment is a major challenge for traffic engineers working in the ITS domain. Finally, additional understanding for automation/control and communication protocols is required in the era of “smart cities.”

Samer Hamdar Department of Civil and Environmental Engineering, George Washington University, Washington, DC, USA [email protected] Alireza Talebpour Department of Civil and Environmental Engineering, University of Illinois, Urbana-Champaign, Urbana, IL, USA Robert Bertini School of Civil and Construction Engineering, College of Engineering, Oregon State University, Corvallis, OR, USA

Acknowledgments

The co-editors in this special issue would like to thank all the participants in the 2017 Traffic and Granular Flow Conference (TGF’17) in general and the authors who accepted the invitation to submit their papers to this TGF Special Issue in particular. Moreover, we are grateful to the anonymous referees for their efforts in reviewing the papers and providing constructive comments. Finally, we extend our thanks to the Editor-in-Chief, Prof. Asad Khattak, and the Journal of Intelligent Transportation Systems (JITS) editorial team for their guidance during the initiation and the publication of this special issue. We are humbled by the patience, the responsiveness, and the understanding of all the contributors (the authors, the reviewers, and the editorial board/team), especially during this COVID-19 Pandemic.

Disclosure statement

No potential conflict of interest was reported by the author(s).