Posted by Yechen Li and Neha Arora, Software Engineers, Google Research
Fifteen minutes. That’s all it took to evacuate the Colosseum, the largest amphitheater in the world that still stands as an engineering marvel. Even after two thousand years, this design continues to effectively move massive crowds out of sports and entertainment venues. However, the challenge lies in navigating the traffic that accumulates in the surrounding streets. This problem, which has persisted for centuries, remains unsolved. In Rome, they tackled this issue by restricting private traffic on the street directly adjacent to the Colosseum. While this approach worked there, what about other locations? What about events like the Superbowl or a Taylor Swift concert?
One potential solution to this problem is the use of simulation models, often referred to as “digital twins”. These models are virtual replicas of real-world transportation networks that aim to capture every detail, from the layout of streets and intersections to the flow of vehicles. Traffic experts utilize these models to alleviate congestion, reduce accidents, and enhance the experience of drivers, riders, and pedestrians. Our team has previously employed these models to assess the sustainability impact of routing, test evacuation plans, and showcase simulated traffic in Maps Immersive View.
Calibrating high-resolution traffic simulations to match the specific dynamics of a particular setting has long been a challenge in this field. However, recent advancements in transportation science, such as understanding the relationship between segment demands and speeds for road segments with traffic signals, along with the availability of aggregate mobility data and detailed Google Maps road network data, have paved the way for efficient optimization on a global scale.
To test this technology in a real-world setting, Google Research collaborated with the Seattle Department of Transportation (SDOT) to develop simulation-based traffic guidance plans. Our objective was to assist thousands of attendees at major sports and entertainment events in quickly and safely exiting the stadium area. The proposed plan successfully reduced average travel times by 7 minutes for vehicles leaving the stadium region during large events. We implemented this plan in collaboration with SDOT using Dynamic Message Signs (DMS) and verified its impact over multiple events between August and November 2023. One of our policy recommendations was to divert traffic from S Spokane St, a major road that connects the area to highways I-5 and SR 99, which often experiences congestion after events. The suggested changes improved traffic flow through highways and arterial streets near the stadium and reduced the length of vehicle queues behind traffic signals.
For this project, we developed a new simulation model of the area surrounding Seattle’s stadiums. The goal of this model was to accurately replicate each traffic situation for a specific day. We utilized an open-source simulation software called Simulation of Urban MObility (SUMO), which offers behavioral models to describe traffic dynamics, including driver decision-making processes such as car-following, lane-changing, and speed limit compliance. Additionally, insights from Google Maps helped define the network’s structure and various static segment attributes, such as the number of lanes, speed limits, and presence of traffic lights.
Travel demand plays a crucial role in the simulation and is computed by decomposing the road network of a metropolitan area into zones. We define travel demand as the expected number of trips between origin and destination zones within a given time period and represent it as aggregated origin-destination (OD) matrices. To determine the initial expected number of trips, we utilize aggregated and anonymized mobility statistics. We then solve the OD calibration problem by combining this initial demand with observed traffic statistics, such as segment speeds, travel times, and vehicular counts, to recreate event scenarios. By modeling traffic patterns during multiple past events at Seattle’s T-Mobile Park and Lumen Field, we can evaluate the accuracy of our simulation by comparing it to aggregated and anonymized traffic statistics. Analyzing these event scenarios allows us to understand the impact of different routing policies on congestion in the region.
After working closely with SDOT and the Seattle Police Department (SPD) to gain local knowledge, we identified the most congested routes that required improvement. These routes included traffic from T-Mobile Park stadium parking lot’s Edgar Martinez Dr. S exit to eastbound I-5 highway/westbound SR 99 highway, traffic through Lumen Field stadium parking lot to northbound Cherry St. I-5 on-ramp, and traffic going southbound through Seattle’s SODO neighborhood to S Spokane St. We developed routing policies and evaluated them using the simulation model. To expedite traffic dispersal, we tested policies that redirected northbound/southbound traffic from the nearest ramps to further highway ramps, reducing wait times. We also experimented with opening HOV lanes to event traffic, suggesting alternate routes (e.g., SR 99), and implementing load sharing between different lanes to reach the nearest stadium ramps.
We conducted simulations for multiple events with varying traffic conditions, event times, and attendee counts. For each policy, the simulation replicated post-game traffic and provided the travel time for vehicles from leaving the stadium to reaching their destination or leaving the Seattle SODO area. The time savings were calculated as the difference in travel time before and after the policy and are shown in the table below for small and large events, considering different percentages of affected vehicles (10%, 30%, or 50%).
Based on these simulation results, feasibility of implementation, and other considerations, SDOT decided to implement the “Northbound Cherry St ramp” and “Southbound S Spokane St ramp” policies using DMS during large events. The signs suggest alternative routes to drivers to reach their destinations. By rerouting 30% of traffic during large events, the combination of these two policies leads to an average travel time savings of 7 minutes per vehicle.
This project exemplifies the power of simulations in modeling, identifying, and quantifying the impact of proposed traffic guidance policies. Simulations enable network planners to identify underutilized segments and evaluate the effects of different routing policies, ultimately leading to better traffic distribution. Our offline modeling and online testing demonstrate that our approach can reduce total travel time. Further enhancements can be made by incorporating additional traffic management strategies, such as optimizing traffic lights. While simulation models have historically been time-consuming and affordable only for large cities and high-stakes projects, our investment in scalable techniques aims to make these models accessible to more cities and use cases worldwide.
Acknowledgements:
We would like to express our gratitude to Alex Shashko, Andrew Tomkins, Ashley Carrick, Carolina Osorio, Chao Zhang, Damien Pierce, Iveel Tsogsuren, Sheila de Guia, and Yi-fan Chen for their contributions to this project. Special thanks to John Guilyard for the visual design. We also extend our thanks to our partners at SDOT, including Carter Danne, Chun Kwan, Ethan Bancroft, Jason Cambridge, Laura Wojcicki, Michael Minor, Mohammed Said, and Trevor Partap, as well as our partners at SPD, Lt. Bryan Clenna, and Sgt. Brian Kokesh.
Source link
