A Nature Cities paper reports a large-scale field experiment testing whether navigation platforms can be used to reduce urban congestion. Over six months in 10 major US cities, a modified Google Maps routing algorithm gently redirected trips away from preselected recurring bottlenecks. Although fewer than 2% of observed trips received altered directions, the coordinated intervention produced measurable improvements in speeds, fuel consumption, and CO2e emissions across the urban networks.
Why this matters
Vehicle mobility is central to modern life but costly: drivers spend on average 2.6 years of their lives driving, and passenger cars and vans account for roughly 10% of global CO2 emissions. Improving how transportation networks are used can therefore yield both time‑saving and climate benefits.
Prior work and motivation
The growing presence of connected vehicles, navigation services, smart‑city infrastructure, and autonomous vehicles creates opportunities to both measure and optimize transportation systems. Google Research’s Project Green Light previously demonstrated the effectiveness of infrastructure‑level AI interventions for traffic signals. However, system‑wide route optimization for ground transport has lacked large‑scale empirical validation.
Experimental design
- Location: 10 major US cities, selected for congestion patterns and availability of ground truth data.
- Duration and design: a six‑month city‑wide switchback (crossover) experiment that alternated between the treatment (modified routing) and control (unaltered routing) on consecutive days to measure effects robustly.
- Intervention: the Google Maps algorithm was adjusted to prefer alternative routes with similar travel times and segment types, steering trips away from about 100 historically congested road segments per city.
- Reach: under 2% of observed trips received altered routing recommendations as a result of the experiment.
Analysis method
The team used a hierarchical Bayesian outcome modeling framework that jointly models aggregate city‑level and localized hourly parameters. This hierarchical approach permits information sharing across cities and times, improving estimate stability for subgroups.
Key results
- On targeted segments: across cities the median increase in driving speeds was about 2%, corresponding to a median reduction in fuel consumption of roughly 0.5–1.0% on those segments.
- Across all affected segments (including those receiving redirected traffic and those offloaded): the median speed increase was about 0.35%, rising to roughly 0.5% during morning and afternoon peak hours.
- Emissions: at the scale and energy demands of the studied cities, these changes correspond to potential savings of thousands of tons of CO2e per city per year.
Improvements in speeds and emission rates were both widespread and statistically significant. The mechanism was strategic diversion from major bottlenecks: by dispersing vehicles more efficiently, peripheral roads were able to maintain higher average speeds and lower emissions even while absorbing greater volumes.
Practical implications
The results show that network‑aware navigation can proactively shape traffic flow for broad societal benefit. Coordinating a small fraction of trips yields systemic gains that accrue to all road users, not solely to users of a particular app. Moreover, the study provides a repeatable experimental blueprint for evolving from individual trip optimization toward cooperative routing and other system‑level traffic management strategies, such as dynamic signal control and real‑time network optimization, as smart‑city infrastructure advances.
Conclusion
Relatively simple, coordinated rerouting implemented through navigation platforms can produce measurable reductions in congestion, fuel use, and emissions at city scale. This work establishes both empirical evidence and an experimentation framework for further research and deployment of network‑aware traffic management.
Acknowledgements
This work was conducted in collaboration with Alexandre Bayen, Andrew Tomkins, Theophile Cabannes, Kevin Chen, Yechen Li, Marc Nunkesser, Prem Ramaswami, Eray Turkel, Shoshana Vasserman, and Haizheng Zhang.
