Dissipation of stop-and-go waves via control of autonomous vehicles: Field experiments

https://doi.org/10.1016/j.trc.2018.02.005Get rights and content

Highlights

  • Experiments on a circular track with 20+ vehicles show stop-and-go waves emerge.

  • Control of an autonomous vehicle can dampen stop-and-go waves in field experiments.

  • Control of one autonomous vehicle reduces total traffic fuel consumption.

  • Mobile traffic control is possible when a small fraction of vehicles are automated.

Abstract

Traffic waves are phenomena that emerge when the vehicular density exceeds a critical threshold. Considering the presence of increasingly automated vehicles in the traffic stream, a number of research activities have focused on the influence of automated vehicles on the bulk traffic flow. In the present article, we demonstrate experimentally that intelligent control of an autonomous vehicle is able to dampen stop-and-go waves that can arise even in the absence of geometric or lane changing triggers. Precisely, our experiments on a circular track with more than 20 vehicles show that traffic waves emerge consistently, and that they can be dampened by controlling the velocity of a single vehicle in the flow. We compare metrics for velocity, braking events, and fuel economy across experiments. These experimental findings suggest a paradigm shift in traffic management: flow control will be possible via a few mobile actuators (less than 5%) long before a majority of vehicles have autonomous capabilities.

Introduction

The dynamics of traffic flow include instabilities as density increases, where small perturbations amplify and grow into stop-and-go waves that travel backwards along the road (Treiterer and Myers, 1974, Sugiyama et al., 2008, Flynn et al., 2009, Kerner, 2012). These so-called phantom traffic jams are an experimentally reproducible phenomenon, as demonstrated in different experiments (Sugiyama et al., 2008, Tadaki et al., 2013, Jiang et al., 2014, Jiang et al., 2017). Common wave triggers include lane changing (Laval and Daganzo, 2006, Laval, 2006, Zheng et al., 2011), but they can even be generated in the absence of any lane changes, bottlenecks, merges, or changes in grade (Sugiyama et al., 2008, Tadaki et al., 2013). Moreover, these waves can be captured in microscopic models of individual vehicle motion (Bando et al., 1995, Nagel and Schreckenberg, 1992, Garavello et al., 2016) (see also the reviews (Brackstone and McDonald, 1999, Chowdhury et al., 2000, Helbing, 2001)) and macroscopic models described via solutions to continuum problems (Flynn et al., 2009, Payne, 1971, Whitham, 1974, Aw and Rascle, 2000, Zhang, 2002, Greenberg, 2004). Since these waves emerge from the collective dynamics of the drivers on the road, they are in principle avoidable if one could affect the way people drive. Recognizing the rapid technological innovations in traffic state estimation and control, this work provides experimental evidence that these waves can be reduced by controlling a small number of vehicles in the traffic stream.

A necessary precursor to dissipating traffic waves is to detect them in real-time. Advancements in traffic state estimation (Gazis and Knapp, 1971, Wang and Papageorgiou, 2005, Blandin et al., 2012) have facilitated high resolution traffic monitoring, through the advent of GPS smartphone sensors (Herrera et al., 2010, Work et al., 2010, Fabritiis et al., 2008, Hofleitner et al., 2012) that are part of the flow—termed Lagrangian or mobile sensors. Now commercialized by several major navigation services, the use of a small number of GPS equipped vehicles in the traffic stream has dramatically changed how traffic is monitored for consumer-facing mobility services, which previously relied on predominantly fixed sensing infrastructure.

Currently, traffic control is dominated by control strategies that rely on actuators at fixed locations or are centralized. Such systems include variable speed advisory (VSA) or variable speed limits (VSL) (Nissan and Koutsopoulos, 2011, Hegyi et al., 2005b, Smulders, 1990, Hegyi et al., 2008, Popov et al., 2008), which are commonly implemented through signs on overhead gantries, and ramp metering (Papageorgiou and Kotsialos, 2002, Gomes and Horowitz, 2006, Papageorgiou et al., 1991), which relies on traffic signals on freeway entrance ramps. More recently, coordinated systems to integrate both ramp metering and variable speed limits have been proposed (Hegyi et al., 2005a, Papamichail et al., 2008, Lu et al., 2010, Han et al., 2017b). A common challenge of VSL and ramp metering systems is the small flexibility of the systems due to the high cost of installation of the fixed infrastructure, which consequently limits the spatial resolution of the control input. Additionally, compliance with the speed advisory is not guaranteed, which can limit the effectiveness of the control strategy.

Recent advancements in vehicular automation and communication technologies have the potential to substantially change surface transportation (Wadud et al., 2016, Harper et al., 2016, Milakis et al., 2017, Wan et al., 2016, Wang et al., 2017). In particular, these advancements provide new possibilities and opportunities for traffic control in which these smart vehicles act as Lagrangian actuators of the bulk traffic steam. When a series of adjacent vehicles on a roadway are connected and automated, it is possible to form dense platoons of vehicles which leave very small gaps. A key challenge for vehicle platoons is to design control laws in which the vehicle platoon remains stable, for which significant theoretical and practical progress has been made (Levine and Athans, 1966, Swaroop and Hedrick, 1996, Shladover, 1995, Fenton and Mayhan, 1991, Darbha and Rajagopal, 1999, Besselink and Johansson, 2017, Ioannou et al., 1993, Buehler et al., 2009, Rajamani et al., 1998). Recent work has shown that commercially-implemented, string-stable adaptive cruise control (ACC) systems may result in an unstable traffic state when implemented on a platoon of ACC-enabled vehicles, motivating the need for vehicle connectivity in such systems (Milanés and Shladover, 2014, Milanés et al., 2014). In contrast to the vehicle platoon setting, in which all vehicles are controlled, or the variable speed limit and ramp metering strategies which actuate the flow at fixed locations, this research aims to dissipate congestion-based stop-and-go traffic waves using only a sparse number of autonomous vehicles already in the flow, without changing how the other, human-driven, vehicles operate.

The notion to dissipate stop-and-go waves via controlling vehicles in the stream represents a shift from stationary to Lagrangian control, mirroring the transition to Lagrangian sensing that has already occurred. The key advantage in mobile sensing projects (Herrera et al., 2010, Fabritiis et al., 2008, Hofleitner et al., 2012) is that a very small number of vehicles being measured (3–5%) suffices to estimate the traffic state on large road networks (Work et al., 2010). In the same spirit, our research experimentally demonstrates that a small number of Lagrangian controllers suffices to dampen traffic waves.

The ability of connected and automated vehicles to change the properties of the bulk traffic flow is already recognized in the transportation engineering community. For example, the works (Davis, 2004, Talebpour and Mahmassani, 2016, Guériau et al., 2016, Wang et al., 2016b, Knorr et al., 2012) directly address the setting where a subset of the vehicles are equipped with automated and/or connected technologies, and then assess via a stability analysis or simulation the extent to which the total vehicular flow can be smoothed. Recently, several works have explored extensions to the variable speed limit control strategies in which connected or automated vehicles are used to actuate the traffic flow (Wang et al., 2016a). For example, the work by Han et al. (2017a) develops a VSL strategy that is implemented in simulation with connected vehicles where the traffic evolves according to the kinematic wave theory. It follows a similar strategy proposed by van de Weg et al. (2014), where a coordinated VSL and ramp metering strategy is implemented via actuation of the entire vehicle fleet (i.e., 100% penetration rate). Although not explicitly designed as a variable speed limit controller, Nishi et al. (2013) advocates a “slow-in, fast-out” driving strategy to eliminate traffic jams, using a microscopic model also in line with kinematic wave theory. The work by He et al. (2016) proposes a similar jam absorbing strategy as Nishi et al. (2013) based on Newell’s car following theory, and its effectiveness is assessed in simulation.

Interestingly, an experimental test of the “slow-in, fast-out” strategy (Nishi et al., 2013) is provided by Taniguchi et al. (2015), in which five vehicles are driven on a closed course. The lead vehicle in the platoon of five vehicles drives initially at a constant speed, then decelerates as if driving through a congestion wave, and then accelerates back to the cruising speed. The third vehicle in the platoon initially leaves a large gap, and due to the extra gap it is able to maintain the cruising speed and effectively absorb the jam. In contrast to the experiment by Taniguchi et al. (2015), the present work fully replicates the setup of Sugiyama et al. (2008) and Tadaki et al. (2013), in which the stop-and-go wave is generated naturally from the human drivers in the experiment, without an external cause. Moreover, the controllers proposed in the present work are distinct.

We also note some preliminary field experiments to harmonize speeds via connected vehicles are recently reported by Ma et al. (2016) and Lu et al. (2015), in part to measure the impact of connected vehicles following an infrastructure-generated advisory speed on the traffic stream behind the connected vehicles. In the present article, we instead dampen waves on a closed ring, which simplifies the experimental setup, and facilitates detailed data collection on the performance of the controllers.

The present article is inspired by the work of Sugiyama et al. (2008) and Tadaki et al. (2013) which are the first works to demonstrate via experiments that traffic waves can emerge as a result of human driving behavior alone. A series of experiments is conducted where approximately 20 vehicles drive in a ring of fixed radius with each driver following the vehicle in front of them. The experiments of Sugiyama et al. (2008) and Tadaki et al. (2013) are foundational because they demonstrate the emergence of traffic waves caused (unintentionally) by human driving behavior. However, they do not offer a solution for dampening these waves.

To address this gap, we design and execute a series of ring-road experiments which show that an intelligently controlled autonomous vehicle is able to dampen human-generated stop-and-go waves. The experimental setup (described in Section 2) follows the setting of Sugiyama et al. (2008) and Tadaki et al. (2013), with the modification that one vehicle is an autonomous-capable vehicle which can run a variety of longitudinal control laws. Similar to the Sugiyama et al. (2008) experiment, the position and velocity of each vehicle is tracked via a 360 degree camera. We additionally instrument each vehicle in the 22-car fleet (see Table 5 in Appendix A.1 for a detailed description of the vehicle fleet) with an OBD-II scanner to log the real-time fuel consumption of each vehicle, such that the impact of the traffic waves and controllers on the bulk fuel consumption can be recorded.

The experimental setup used allows for us to isolate the effect traffic instabilities caused by human car following behavior, while eliminating other sources of congestion. Specifically, this work does not aim to quantify the effect of AVs on congestion triggers such as lane changing or geometric bottlenecks such as a reduction in lanes. Instead, the experimental setup is designed to easily study traffic waves caused by the car-following behavior of human drivers, and consequently follows a similar experimental design used in Sugiyama et al. (2008). It is important to stress that the ring setup does not represent all of the complexities of human driving behavior on long stretches of roadway. However it does allow the emergent phenomenon of stop-and-go waves, which are observed in freeway traffic flows, to be reliably reproduced so that the effectiveness of AV control laws designed to dampen these waves can be quantified. These experiments should be viewed as a precursor to larger field experiments on real freeways.

We present three experiments (labeled as A, B, and C in Section 4) and two distinct control strategies (detailed in Section 3) that can be used to dampen stop-and-go waves created by human drivers. The first control strategy is to follow a fixed average velocity (selected based on observation) as closely as possible without collisions. It is implemented in Experiment A via an automatic control algorithm (called FollowerStopper) and in Experiment B via a carefully trained human driver. The second type of control strategy is a proportional-integral (PI) controller with saturation, which is a natural extension of the PI controller, a simple and widely used controller in industrial applications. The controller is only based on the knowledge of the autonomous vehicle speed over a time horizon. The control action is saturated at small gaps to avoid collisions, and long gaps to avoid slowing down of traffic. Compared to the average velocity controllers (Experiments A and B), the PI controller with saturation directly estimates the average velocity and thus needs no external input.

The results of each of the three experiments are presented and compared in Section 4. In each experiment, stop-and-go waves arise dynamically when all vehicles are under human control. Once one vehicle is activated to be autonomous (with the control algorithms described in Section 3), the traffic waves are dissipated. Compared to when waves are present, the Lagrangian control results in up to 40% less fuel consumption, and a throughput increase of up to 15%. Future perspectives for Lagrangian vehicular control are provided in Section 5.

Section snippets

Experimental methodology

We briefly describe the experimental setting in which stop-and-go waves are observed to develop and subsequently dampened via control of a single vehicle in the experiment (mimicking a uniform low penetration rate on a long freeway stretch). The experiments follow the ring setting of Sugiyama et al. (2008) and Tadaki et al. (2013). A key advantage of the ring road experimental setup (Sugiyama et al., 2008, Tadaki et al., 2013) is that it removes other effects like boundary conditions, merging

Description of controllers of the autonomous vehicle

This section presents the velocity controllers implemented on the CAT Vehicle, which are observed to dampen traffic waves on the ring track. The controllers are broadly motivated by the fact that mathematical models of vehicular traffic can be stabilized via the control of a small number of vehicles, see for example (Davis, 2004, Talebpour and Mahmassani, 2016, Guériau et al., 2016, Wang et al., 2016b, Cui et al., 2017). The main goal when stabilizing the traffic flow is to control the traffic

Experimental results: Dampening traffic waves with a single vehicle

The experimental results are presented in this section. To be able to effectively compare the results of the experiments, it is important to define metrics, which are consistent across the experiments. To this end, we present the metrics used to describe the traffic flow in Section 4.1. With the metrics fully defined, we present the results of the three experiments conducted in Section 4.2.

Conclusions

AVs can revolutionize the control of traffic flow. They offer the potential to shift from localized control measures, like ramp metering, and centralized ones, as variable speed limit gantries, to Lagrangian actuators immersed in the traffic stream. Strikingly, it is not necessary for all vehicles to be automated in order to benefit from mobile actuation. A single autonomous vehicle can control the flow of at least 20 human-controlled vehicles around it, with substantial reductions in velocity

Acknowledgements

This material is based upon work supported by the U.S. National Science Foundation under Grant No. CNS-1446715 (B.P.), CNS-1446690 (B.S.), CNS-1446435 (J.S.), and CNS-1446702 (D.W.). The authors thank the University of Arizona Motor Pool in providing the vehicle fleet. They offer additional special thanks for the services of N. Emptage in carrying out the experiment logistics.

References (82)

  • J.C. Herrera et al.

    Evaluation of traffic data obtained via GPS-enabled mobile phones: The mobile century field experiment

    Transport. Res. Part C: Emerging Technol.

    (2010)
  • J.A. Laval et al.

    Lane-changing in traffic streams

    Transport. Res. Part B: Methodol.

    (2006)
  • D. Milakis et al.

    Policy and society related implications of automated driving: a review of literature and directions for future research

    J. Intell. Transport. Syst.

    (2017)
  • V. Milanés et al.

    Modeling cooperative and autonomous adaptive cruise control dynamic responses using experimental data

    Transport. Res. Part C: Emerging Technol.

    (2014)
  • R. Nishi et al.

    Theory of jam-absorption driving

    Transport. Res. Part B: Methodol.

    (2013)
  • A. Nissan et al.

    Evaluation of the impact of advisory variable speed limits on motorway capacity and level of service

    Procedia-Social Behav. Sci.

    (2011)
  • I. Papamichail et al.

    Integrated ramp metering and variable speed limit control of motorway traffic flow

    IFAC Proc. Vol.

    (2008)
  • S. Smulders

    Control of freeway traffic flow by variable speed signs

    Transport. Res. Part B: Methodol.

    (1990)
  • A. Talebpour et al.

    Influence of connected and autonomous vehicles on traffic flow stability and throughput

    Transport. Res. Part C: Emerging Technol.

    (2016)
  • Z. Wadud et al.

    Help or hindrance? the travel, energy and carbon impacts of highly automated vehicles

    Transport. Res. Part A: Policy Practice

    (2016)
  • N. Wan et al.

    Optimal speed advisory for connected vehicles in arterial roads and the impact on mixed traffic

    Transport. Res. Part C: Emerging Technol.

    (2016)
  • M. Wang et al.

    Connected variable speed limits control and car-following control with vehicle-infrastructure communication to resolve stop-and-go waves

    J. Intell. Transport. Syst.

    (2016)
  • R. Wang et al.

    Comparing traffic state estimators for mixed human and automated traffic flows

    Transport. Res. Part C: Emerging Technol.

    (2017)
  • Y. Wang et al.

    Real-time freeway traffic state estimation based on extended Kalman filter: a general approach

    Transport. Res. Part B: Methodol.

    (2005)
  • H.M. Zhang

    A non-equilibrium traffic model devoid of gas-like behavior

    Transp. Res. B

    (2002)
  • Z. Zheng et al.

    Applications of wavelet transform for analysis of freeway traffic: Bottlenecks, transient traffic, and traffic oscillations

    Transport. Res. Part B: Methodol.

    (2011)
  • Åström, K.J., Hägglund, T., 2005. Advanced PID Control. ISA, Research Triangle Park,...
  • Åström, K.J., Murray, R.M., 2008. Feedback systems: An introduction for scientists and...
  • A. Aw et al.

    Resurrection of second order models of traffic flow

    SIAM J. Appl. Math.

    (2000)
  • M. Bando et al.

    Dynamical model of traffic congestion and numerical simulation

    Phys. Rev. E

    (1995)
  • A. Berdahl et al.

    Emergent sensing of complex environments by mobile animal groups

    Science

    (2013)
  • Besselink, B., Johansson, K.H., 2017. String stability and a delay-based spacing policy for vehicle platoons subject to...
  • M. Buehler et al.
    (2009)
  • Cui, S., Seibold, B., Stern, R., Work, D.B., 2017. Stabilizing traffic flow via a single autonomous vehicle:...
  • L. Davis

    Effect of adaptive cruise control systems on traffic flow

    Phys. Rev. E

    (2004)
  • De Fabritiis, C., Ragona, R., Valenti, G., 2008. Traffic estimation and prediction based on real time floating car...
  • J.E.A. Dias et al.

    Longitudinal model identification and velocity control of an autonomous car

    IEEE Trans. Intell. Transport. Syst.

    (2015)
  • Environmental Protection Agency, 1995. Control of air pollution from new motor vehicles and new motor vehicle engines;...
  • R.E. Fenton et al.

    Automated highway studies at the Ohio State University-an overview

    IEEE Trans. Vehicular Technol.

    (1991)
  • M.R. Flynn et al.

    Self-sustained nonlinear waves in traffic flow

    Phys. Rev. E

    (2009)
  • Garavello, M., Han, K., Piccoli, B., 2016. Models for Vehicular Traffic on Networks. American Institute of Mathematical...
  • Cited by (625)

    • Automatic vehicle trajectory data reconstruction at scale

      2024, Transportation Research Part C: Emerging Technologies
    View all citing articles on Scopus
    1

    R. Stern, S. Cui, and M.L. Delle Monache contributed equally to this work.

    2

    B. Piccoli, B. Seibold, J. Sprinkle, and D. Work contributed equally to this work.

    View full text