Source-linked AI summary
Dissipation of stop-and-go waves via control of autonomous vehicles: Field experiments
Raphael E. Stern, Shumo Cui, Maria Laura Delle Monache, Rahul Bhadani, Matt Bunting, Miles Churchill, Nathaniel Hamilton, R'mani Haulcy, Hannah Pohlmann, Fangyu Wu, Benedetto Piccoli, Benjamin Seibold, Jonathan Sprinkle, Daniel B. Work
TL;DR
Stop-and-go waves can emerge without infrastructure bottlenecks or lane changes, but prior experiments did not provide a way to dampen them. This study uses ring-road experiments with one controlled autonomous vehicle among human-driven vehicles and finds that simple control strategies reduce wave-related speed variability, braking, and fuel consumption.
Problem
Prior ring-road experiments showed that human driving can generate traffic waves without bottlenecks or lane changes, but did not offer a solution for dampening them.
Method
The study conducts three experiments with a 22-car fleet, tracking vehicle motion and fuel consumption while applying FollowerStopper or saturated PI control to one autonomous-capable vehicle.
Results
54.7% reduction in speed variability, 27.9% lower fuel consumption, and 74.4% lower excessive-braking rate were observed with the saturated PI controller, alongside a 2.5% throughput reduction.
Takeaways & Limitations
A single autonomous vehicle can control the flow of at least 20 human-controlled vehicles, suggesting that sparse mobile actuators can influence traffic dynamics without dedicated actuation infrastructure.
Takeaways & Limitations
The experiments represent a single-lane roadway, so the benefits of Lagrangian actuators on multi-lane urban freeways require future experiments.
Abstract
from arXiv · showhide
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.
1. Introduction
Stop-and-go waves can emerge without bottlenecks or lane changes, but earlier ring-road experiments did not provide a damping solution. This study addresses that gap by testing whether one intelligently controlled autonomous vehicle can dampen waves and improve traffic-flow outcomes.
- Motivation: Fixed-location and centralized controls, including variable speed limits and ramp metering, have limited spatial flexibility because of infrastructure costs.Their effects are also limited between fixed control points.
- Contribution: Lagrangian control shifts traffic actuation from stationary infrastructure to vehicles moving within the traffic stream.The approach parallels mobile traffic sensing, where 3–5% of vehicles can suffice to estimate traffic state across large networks.
- Motivation: Stop-and-go waves can arise from human driving behavior even without infrastructure bottlenecks or lane changes.Earlier experiments with approximately 20 vehicles demonstrated reproducible wave emergence but did not offer a damping solution.
- Approach: The study uses ring-road experiments with one autonomous-capable vehicle, tracking vehicle motion and logging fleet-wide fuel consumption.The setup follows earlier ring experiments while adding autonomous longitudinal control and OBD-II measurements across the 22-car fleet.
- Approach: Three experiments compare two wave-damping strategies: fixed-average-velocity control and a proportional-integral controller with saturation.The PI controller estimates average traffic velocity directly, whereas the average-velocity strategies require an externally selected target.
- Results: Across the experiments, activating one autonomous vehicle dissipates waves, yielding up to 40% less fuel consumption and up to 15% higher throughput.The comparisons are made against periods when waves are present under human control.
2. Experimental methodology
The experiments used a controlled single-lane circular track with 21–22 vehicles, one self-driving-capable vehicle, centralized video tracking, and onboard data logging. Human drivers first generated unsteady traffic, after which the CAT Vehicle was switched into control mode.
- Track and fleet: The study used a 260-meter single-lane circular track with 21–22 vehicles and no major roadside obstacles.The ring setup removes boundary conditions, merging lanes, and intersections from the experiment.
- Track and fleet: One of the 22 passenger vehicles was the self-driving-capable CAT Vehicle, while a trained human driver continuously controlled its steering.Only this vehicle was controlled to dampen the traffic wave; the remaining vehicles were human-driven.
- Data collection: A 360-degree central camera tracked vehicle positions and velocities, while OBD-II loggers recorded real-time fuel consumption.Video processing identified vehicle centers frame by frame to generate smoothed trajectories.
- Procedure: Each experiment lasted 7–10 minutes, began with uniformly spaced vehicles, and allowed waves to persist for at least 45 seconds before control activation.The CAT Vehicle began in manual mode and was switched into a control mode during the experiment.
- Driver instructions: Drivers were instructed to follow the vehicle ahead without passing or colliding, while catching up when gaps opened.The instructions were intended to prevent drivers from intentionally smoothing traffic waves.
3. Description of controllers of the autonomous vehicle
The autonomous vehicle uses two velocity-control strategies to smooth traffic: FollowerStopper commands speeds from gap and relative velocity, while PI control with saturation tracks an estimated average speed. A multi-mode low-level controller converts commanded velocity into accelerator or brake signals with distinct acceleration and braking gains.
- Controller structure: The controllers continuously use the autonomous vehicle’s speed, the gap to the lead vehicle, and the lead vehicle’s estimated speed to determine commanded velocity.The gap is sampled at 30 Hz, and lead speed is estimated from autonomous-vehicle speed plus the gap-rate signal.
- FollowerStopper controller: FollowerStopper commands the desired velocity when safe and lowers it when safety requires, using region boundaries based on deceleration trajectories.The lead velocity is limited to the smaller of the positive lead speed and desired velocity; adaptation regions transition continuously from stopping to safe driving.
- FollowerStopper controller: FollowerStopper divides gap–velocity-difference space into safe, stopping, and adaptation regions.The regions command the desired speed, zero speed, or a transition involving desired and lead-vehicle velocities.
- PI with saturation controller: The PI-with-saturation controller uses a temporal average of the autonomous vehicle’s speed as its desired velocity and adjusts toward it according to the current gap.The averaging interval corresponds to 38 seconds, approximately one lap; saturation makes short-gap behavior follow the lead vehicle and large-gap behavior close the gap.
- PI with saturation controller: The PI controller’s safety distance and gap limits were calibrated through simulation and car-following field tests.The implementation uses lower and upper gap limits of 7 m and 30 m, respectively, and a safety distance based on the 2-second rule with a 4 m minimum.
- Low-level vehicle controls: A multi-mode low-level controller translates commanded velocity into gas and brake signals, using separate gains for acceleration and braking.Each mode is a PID controller whose gains were identified from CAT Vehicle tests, with the vehicle represented by a first-order model.
4. Experimental results: Dampening traffic waves with a single vehicle
Across three ring-road experiments, controlling one autonomous-capable vehicle dampened stop-and-go waves and generally improved velocity variability, fuel use, braking, and sometimes throughput. The strongest outcomes included 80.8% lower velocity standard deviation, 42.5% lower fuel consumption, and 14.1% higher throughput in Experiment A.
- Experiment A: 80.8% lower velocity standard deviation, 42.5% lower fuel consumption, and 14.1% higher throughput were achieved in Experiment A at the best FollowerStopper setting.At 7.50 m/s, excessive braking fell from 8.58 to 0.12 events per vehicle per kilometer.
- Experiment A: In Experiment A, increasing the FollowerStopper set point to 7.50 m/s produced the lowest fuel consumption, 14.13 ℓ/100km, across all experiments.The traffic wave reappeared when the controller was deactivated.
- Experiment C: In Experiment C, wave control reduced standard deviation by 54.7%, fuel consumption by 27.9%, and excessive braking by 74.4%, while throughput fell by 2.5%.The local PI controller used only information measured by the CAT Vehicle and slightly reduced average velocity.
- Conclusion: The experiments demonstrate that low-penetration-rate autonomous-vehicle control can dissipate traffic waves and reduce fuel consumption and braking events.The paper reports throughput increases in some experiments under proper control.
- Cross-experiment comparison: Across experiments, velocity standard deviation fell by 49.5% to 80.8%, fuel consumption fell by 22.1% to 42.5%, and throughput changed by +14.1%, +9.8%, and −2.5%.Excessive braking declined from 8.58–9.66 events/veh/km to 2.47 events/veh/km in the weakest controller and 0.12 events/veh/km in the strongest.
5. Conclusions
The conclusions frame autonomous vehicles as sparse, mobile actuators that can control traffic flow without requiring fleet-wide automation. The experiments also identify simple control structures and a single-lane setting that bounds the demonstrated scope.
- 5. Conclusions: A single autonomous vehicle can control the flow of at least 20 human-controlled vehicles as a mobile actuator.This shifts traffic control from fixed or centralized interventions toward Lagrangian actuation within the traffic stream.
- 5. Conclusions: Simple controllers using the autonomous vehicle’s velocity, spatial gap, and estimated average traffic velocity can dampen traffic waves.The experiments include a fully automatic PI controller with saturation and externally informed FollowerStopper and human-implemented controllers.
- 5. Conclusions: Sparse Lagrangian actuators offer direct control of traffic-flow dynamics without dedicated fixed-location actuation infrastructure.They are contrasted with centralized strategies such as ramp metering, variable speed limits, and traffic-light controls.
- 5. Conclusions: The experiments represent a single-lane roadway, while extending the theory to multi-lane freeways introduces lane-changing and gap-management challenges.The authors state that future multi-lane experiments are needed to fully quantify freeway benefits.
- 5. Conclusions: The control concept could in principle be implemented with existing technology rather than requiring a far-future deployment.The proposed implementation pathway includes adaptive cruise control, intelligent infrastructure, and connected vehicles for external inputs.
Appendix
The appendix describes the vehicle fleet and the procedure for detecting traffic-wave onset. It uses instantaneous velocity statistics and a fixed standard-deviation threshold to identify when waves first appear.
- Appendix: Experiments A and B use vehicles 1 through 21, while Experiment C also includes vehicle 22.The appendix references a full table of vehicle year, make, model, length, and EPA-rated fuel consumption.
- Appendix: At each timestep, the analysis computes the mean and sample standard deviation of all vehicle velocities.The resulting instantaneous velocity statistics are used to characterize oscillatory traffic behavior.
- Appendix: A traffic wave is defined when instantaneous vehicle-velocity standard deviation exceeds 2.5 m/s.This threshold determines the time at which waves first appear in the experiments.
- Appendix: Figures 6a–6c plot instantaneous velocity standard deviation over time for Experiments A, B, and C against the wave threshold.The threshold is shown as a red dashed line.