Rapidly growing cities confront escalating urban transportation challenges, including severe traffic gridlock, significant environmental impacts, and alarmingly high road fatalities, with 38,585 U.S. traffic deaths recorded in 2023 alone. These issues intensify with evolving trends: the surge in ride-sharing services alters parking demands and necessitates efficient traffic management; the rise of e-commerce reshapes urban logistics, requiring innovative curb space and road capacity solutions; and the advent of autonomous vehicles introduces both infrastructure and regulatory hurdles alongside opportunities for enhanced safety and efficiency. Simultaneously, advancements in AI offer powerful capabilities to address these complex problems at scale, providing novel methods for congestion management, public and multimodal transportation enhancement, freight movement support, and community resilience building. In direct response to these dynamic shifts, Google Research proudly introduces Mobility AI, a program dedicated to driving scientific breakthroughs in mobility. This initiative translates cutting-edge research into core technologies accessible through Google Public Sector and Google Maps Platform, equipping transportation leaders with essential tools for data-driven policy formulation, effective traffic management, and continuous oversight of transportation systems, encompassing roads, transit, cycling, and pedestrian infrastructure. This exploration details the foundational research propelling Mobility AI and highlights key technological advancements from Google across the fundamental pillars of measurement, simulation, and optimization, aiming to foster a deeper understanding and replication of transportation networks and mobility patterns, thereby enabling superior design and evaluation of interventions.
A foundation of research and innovation
For over two decades, Google Research has pioneered transportation innovation, developing advanced mapping, routing technologies, and traffic prediction algorithms that empower efficient urban navigation. We have committed extensive resources to understanding traffic dynamics, creating simulations for intervention design, and evaluating traffic management strategies. The availability of high-resolution, global-scale mobility data, combined with significant progress in transportation science and simulation calibration, now enables the development of scalable solutions. Building upon this extensive expertise, Mobility AI represents a concentrated effort to consolidate and integrate Google's domain-specific research. We are delivering the next generation of AI-driven transportation technologies by constructing foundational capabilities that provide the essential scientific and technological backbone for public agencies and their partners to precisely and effectively address urban transportation challenges.
The Mobility AI framework: Measurement, simulation, optimization
Our research strategy is structured around three core pillars:
The Mobility AI framework.
1. Measurement: Understanding mobility patterns
Accurately assessing the current state of the transportation network and mobility patterns is the essential first step toward improvement. This process involves collecting and analyzing real-time and historical data from diverse sources to comprehend both present conditions and historical trends, enabling us to track the effects of implemented changes. Machine learning powers estimations and metric computations, while statistical methods rigorously measure impact. Key areas of focus include:
Congestion functions
Mirroring the well-established fundamental diagrams of traffic flow, congestion functions mathematically quantify how increasing vehicle volume elevates congestion and reduces travel speeds, offering critical insights into traffic behavior. Unlike fundamental diagrams, congestion functions are derived from a subset of vehicles, such as floating car data, rather than the entire traffic stream. We have significantly advanced the understanding of congestion formation and propagation through an ML approach that generated city-wide models. These models facilitate robust inference even on roads with limited data and, through analytical formulation, reveal how adjustments to traffic signals influence flow distribution and congestion patterns within urban environments.
Foundational geospatial understanding
We develop innovative frameworks that utilize techniques like self-supervised learning on geospatial data and movement patterns. This approach generates embeddings that capture both local characteristics and broader spatial relationships, thereby enhancing the understanding of mobility patterns and supporting downstream tasks, particularly in data-scarce scenarios or when complementing other data modalities. Collaboration with related Google Research efforts in Geospatial Reasoning, leveraging generative AI and foundation models, is paramount to advancing these capabilities.
Parking insights
Comprehending urban intricacies extends to parking. Building upon our established work in using ML to predict parking difficulty, Mobility AI aims to deliver superior insights for managing parking availability. This is crucial for a wide range of users, including commuters, ride-sharing drivers, commercial delivery vehicles, and the emerging requirements of autonomous vehicles.
Origin–destination travel demand estimation
Origin–destination (OD) travel demand, which maps the starting and ending points of trips such as daily commutes, goods deliveries, or shopping journeys, forms the bedrock of understanding and optimizing mobility. Identifying these patterns is vital as it precisely indicates where the transportation network experiences strain and where infrastructure improvements or service enhancements are most critically needed. We calibrate OD matrices—tables quantifying trips between locations—to accurately replicate observed traffic patterns, thereby providing a spatially comprehensive understanding essential for the planning and optimization of transportation networks.
Performance metrics: Safety, emissions and congestion impact
We leverage aggregated and anonymized Google Maps traffic trends to evaluate the impact of transportation interventions on congestion and develop models to assess safety and emissions impacts. To construct safety metrics efficiently, we move beyond reactive crash data by utilizing hard braking events (HBEs). HBEs demonstrate a strong correlation with crashes and can serve road safety services in pinpointing high-risk locations and forecasting future collision risks. For measuring environmental impact, we have developed AI models in collaboration with the National Renewable Energy Laboratory (NREL) that predict vehicle energy consumption across various powertrain types (gas, diesel, hybrid, electric). This capability powers fuel-efficient routing within Google Maps, which has already contributed to an estimated avoidance of 2.9 million metric tons of greenhouse gas emissions in the U.S. alone—equivalent to removing approximately 650,000 cars from the road annually. This functionality is fundamental for monitoring the climate and health implications associated with transportation choices.
Impact evaluation
Implementing randomized trials to evaluate transportation policy changes is frequently infeasible. To accurately assess the impact of a given change, we must estimate outcomes that would have occurred in its absence. This is achieved by identifying cities or regions with similar mobility patterns to serve as a control group. Our analysis of New York City's congestion pricing program exemplifies this methodology, employing sophisticated statistical techniques like synthetic controls to rigorously estimate the policy's impact and deliver valuable insights for agencies evaluating interventions.
2. Simulation
Within the simulation pillar, we construct high-resolution digital twins—virtual replicas—of transportation systems. These advanced models enable planners and engineers to rigorously test interventions, forecast potential outcomes, and evaluate "what-if" scenarios, ranging from minor signal timing adjustments to major infrastructure projects or emergency response plans, before real-world implementation. Our objective is to expand the scope of traffic simulations beyond localized areas like single highways or arterials to encompass entire metropolitan networks comprising thousands of road segments, while also ensuring rapid deployment across diverse cities and scenarios. Historically, achieving this proved challenging because simulation parameter calibration to match specific local dynamics was dependent on limited traffic sensor data, restricting our ability to observe and replicate a wide spectrum of congestion patterns. Our research has overcome this obstacle by developing methods to calibrate simulations using more comprehensive data, including aggregated and anonymized speeds, counts, and path travel times, integrated into physics-informed traffic models. Furthermore, acknowledging the significant computational demands of detailed simulation calibration, we are actively pursuing improvements in efficiency alongside advancements in scaling and accuracy.
Given that digital twins are utilized for "what-if" analysis to estimate the effects of prospective interventions on the mobility network, it is critical to understand how uncertainty in simulation inputs, such as demand and supply, influences the simulation outputs. We develop algorithms that precisely quantify this input uncertainty, empowering cities to conduct robust "what-if" analyses using these digital twins.
Watch the film
Traffic Simulation API Video | Mobility AI.
3. Optimization
Leveraging insights derived from measurement and simulation, we apply sophisticated algorithms and AI to identify the most effective solutions for transportation networks. This encompasses:
- Intervention design: Crafting data-driven plans tailored to specific challenges. For instance, we collaborated with Seattle DOT on post-event traffic guidance and developed detailed evacuation plans for Mill Valley, identifying network modifications to enhance effectiveness during emergency response scenarios. These strategies also apply to optimizing traffic flow around work zones or managing freight movement.
- Routing strategies: Developing routing approaches that improve overall network efficiency and minimize social costs.
- Traffic signal control: Optimizing signal timing to reduce delays and emissions, exemplified by Project Green Light.
Shaping the future of Urban Mobility, together
Mobility AI underscores Google Research's enduring commitment to applying scientific advancements, including foundational machine learning and algorithms, to foster more efficient, safe, and resilient transportation systems for functional urban areas. By concentrating on robust measurement, predictive simulation, and effective optimization, we aim to deliver the technological capabilities necessary to manage congestion, enhance public transportation access, and improve community well-being through superior routing, mapping, and modeling. We firmly believe that collaboration is indispensable for advancing Mobility AI. We eagerly seek partnerships with transportation agencies, planners, researchers, and mobility providers to translate Mobility AI's research insights into practical applications, leveraging digital infrastructure and enabling enhanced data sharing for maximum positive impact. If you are interested in learning more about these research advancements and discussing the potential for your organization, please complete our interest form to connect with our team.