Mobility, AI assistants, and urban emissions: an insidious triangle

Abstract

Transportation remains a significant contributor to greenhouse gas emissions, with a substantial proportion originating from road transport and passenger travel in particular. Today, the relationship between transportation and urban emissions is even more complex, given the increasingly prevalent role and the pervasiveness of AI-based GPS navigation systems such as Google Maps and TomTom. While these services ofer benefits to individual drivers, they can also exacerbate congestion and increase pollution if too many drivers are directed onto the same route. In this article, we provide two examples from our research group that explore the impact of vehicular transportation and mobility-AI-based applications on urban emissions. By conducting realistic simulations and studying the impact of GPS navigation systems on emissions, we provide insights into the potential for mitigating transportation emissions and developing policies that promote sustainable urban mobility. Our examples demonstrate how vehicle-generated emissions can be reduced and how studying the impact of GPS navigation systems on emissions can lead to unexpected findings. Overall, our analysis suggests that it is crucial to consider the impact of emerging technologies on transportation and emissions, and to develop strategies that promote sustainable mobility while ensuring the optimal use of these tools.

Keywords

• ban emissions
• In Section 2
• we examine the spatial pat-

1. Introduction

Assessing Greenhouse Gas (GHG) emissions and air pollution is essential to mitigating climate change and promoting human health. Among various sources of emissions, transportation ones significantly increased since 1970, with 11.9% of global GHG emissions in 2016 originating from road transport, 60% of which from passenger travel [ 1, 2 ]. Additionally, transportation emissions contribute to non-CO

2 pollutants such as nitrogen oxides, ozone, particulate matter, and volatile organic compounds, significantly impacting climate change and threatening human health [ 1 ]. Achieving Sustainable Development Goals (SDGs) by 2030 requires urgent action towards reducing cities’ per capita environmental impact [ 3 ]. ban emissions and welfare, it is essential to consider the impact of Artificial Intelligence (AI) on human mobilstudying the impact of GPS navigation systems on emissions can reveal unexpected results. By evaluating the impact of mobility and AI-based applications on emissions, we aim to provide insights into the potential for mitigating transportation emissions and developing policies that promote sustainable urban mobility. We find that a few “gross polluters” are responsible for a tremendous amount of emissions in all three cities, and most vehicles emit significantly less. Indeed, the 2. Understanding urban emissions distribution of emissions per vehicle is associated with a Gini coeficient higher than 0.55 for all the cities and Existing methods to measure vehicles’ emissions vary pollutants. The top 10% of gross polluters in Florence, widely in their level of detail and generalizability. Some Rome, and London are responsible for 47.5%, 50.5%, and smpeatthioo-dtesmreployraolnressmolaulltiosanmsbpuletsliomfitveedhigcelneserwaliitzhabhiliigthy s3p8e.5c%tivoeflyt.heThtoetasltuCdOy2alesmo ifinttdesdtdhuatritnhge tdhiestmribounttiho,nrsedue to their sample size. For example, particulate sensors of CO2 emissions per vehicle of Rome and Florence are and portable emissions measurement systems provide well approximated by a truncated power law, while a accurate emissions measurements in real-world driving stretched exponential well approximates London’s disconditions but are limited in scope [ 7, 8, 9 ]. In contrast, tribution. This pattern is consistent for other pollutants studies using odometer readings provide estimates for as well. Similarly, a few “grossly polluted roads” sufer entire regions but lack instantaneous speed and accelera- from a significant quantity of emissions, but most suftion data [ 10, 11, 12, 13 ]. fer significantly less. The distribution of emissions per

GPS traces ofer a trade-of between these two ex- road is associated with a Gini coeficient higher than 0.64 tremes, providing high spatio-temporal resolution and for all the cities and pollutants, and a truncated power the ability to estimate emissions using microscopic mod- law well approximates it. Figure 1 shows the entire road els while covering a representative fraction of the ve- network of Greater London, with the emissions on each hicle fleet [ 14]. Several studies have used GPS traces road normalised by the road length to better highlight to investigate the relationship between emissions and the diferences between the roads. the urban environment, vehicle miles travelled and fuel consumption, trip rates and travel mode choice, and more [15, 16, 17, 18, 19]. Overall, using GPS traces pro- 2.2. Simulation scenarios vides a valuable tool for understanding the impact of We study how electrifying a certain share of vehicles vehicles on the environment and implementing strate- would change the emissions on the three cities’ roads. gies to reduce emissions. In this setting, even if a vehicle’s electrification changes

Despite the variety of literature, it remains unclear its driver’s mobility behaviour, it would not create any what statistical patterns characterize the distribution of emissions. emissions per vehicle and road, how these distributions We find that the electrification of just the top 1% of change over time and space, and how this information gross polluters would reduce emissions as much as electrican be used to simulate emission reduction scenarios. fying 10% of random vehicles. In contrast, the percentage While studies have reported that the distribution of emis- reduction of the overall emissions grows almost linearly sions from on-road remote sensing sites across vehicles when the share of electric vehicles is chosen randomly. is skewed [20, 21], this finding has been questioned due A Generalised Logistic Function (GLF) approximates the to the limitations of this type of measurement [22]. growth rate when the gross polluters are electrified first. The model gives 2 = 0.99 for Rome, and similar results 2.1. Emissions patterns hold for Florence. In Greater London, the growth starts slowly: there are fewer vehicles with high emissions levIn a recent study [23], we analyse anonymous GPS tra- els, and electrifying the most polluting vehicles is slightly jectories describing 423,018 trips from 16,715 private less efective in reducing emissions than in the other two light-duty vehicles moving in Greater London, Rome, cities [23]. and Florence throughout January 2017 to compute vehi- Given the increasing importance of remote workcles’ emissions. The trajectories are produced by onboard ing [25, 26], we also simulate the impact of a massive GPS devices that automatically turn on when the vehicle shift to remote working on reducing vehicle emissions. starts, transmitting a point every minute to the server We assume that this working style eliminates commuting via a GPRS connection. trips. We then perform a simulation in which a grow

We develop a methodological framework to compute ing share of these commuters become home workers, vehicles’ emissions from their raw GPS trajectories, and i.e., they no longer travel between their home and work we use a microscopic emissions model that estimates locations (detected from GPS traces). the vehicles’ instantaneous emissions of carbon dioxide We find that emissions reduction is more efective (CO2), nitrogen oxides (NOx), particulate matter (PM), when the home workers are gross polluters: remote and volatile organic compounds (VOC) from speed, ac- working for the top 1% gross polluters leads to the same celeration, and fuel type. reduction reached if they were ≈ 4% random vehicles.

Again, a GLF fits emissions reduction when the gross a driver’s routing choice cannot be evaluated in isolation polluters become home workers. Overall, these results because it depends on the concurrent choices of other demonstrate that targeting specific profiles of vehicles drivers in the city. If too many drivers select the same can significantly improve emission reduction policies. ”eco-friendly” route, the route may become congested, reducing its eco-friendliness. Therefore, a better understanding of the impact of individual routing choices on 3. Understanding impact of AI the urban environment is necessary. In a recent paper [31], we designed a simulative frameAccording to preliminary research, the influence of navi- work – TrafiCO2 – using the state-of-the-art trafic simgation applications on the urban environment is a topic ulator SUMO [32] to create realistic simulations of trafic that remains unclear and incomplete, as existing stud- under diferent settings. The simulations were conducted ies produce inconsistent and sporadic outcomes [27, 28]. in the city of Milan assuming that vehicles would either On the one hand, these apps may contribute to miti- adhere to the directions of commercial navigation sysgating CO2 emissions [29]. On the other hand, their tems – OpenStreetMap (OSM) and TomTom (TT) – or usage may increase population exposure to pollution follow a randomised deviation from the fastest route that in highly populated regions [30]. Real-time navigation emulates the unpredictability and irrationality of human apps provide drivers with optimal routes to reach their drivers. destinations, considering the current trafic conditions. We varied the percentage of vehicles of the fleet circuHowever, despite their undeniable practicality, online lating in Milan that followed a navigation app’s suggesnavigation applications can in principle generate several tions, in order to assess the impact of the rate of routed issues in urban trafic [ 5, 4 ]. These apps are usually opti- vehicles on the urban environment. We found that the mized to minimize individual drivers’ travel time with- greater the number of vehicles following the navigation oduritvceorsn’sbideehraivnigouthreocnotllheectciivtey’ismopvaecrtaollfttrhaefic saigtguraetgioante.d tahpep’csitsyu:ggbelisntidolny, ftohlelohwiginhgertthhee rteoctoalmCmOe2nedmatiisosniosnosfina For example, these apps may not factor in whether the navigation app, which are optimised from an individual recommended routes could handle the additional trafic standpoint, can lead to trafic congestion in some areas generated by the app or whether this trafic could pose a of the city, thus leading to spatial polarisation which rerisk to safety or lead to further pollution. The impact of sults in increased travel time and emissions. Conversely, taking “noisy” routes increases the diversity of travelled paths, resulting in a better distribution of trafic on the road network and a decreased travel time and emissions.

The fraction of routed vehicles also influenced the spatial distribution of emissions in the city, with more emissions concentrating on the external ring road when more vehicles were routed.

Our results also suggested that introducing randomness into the path generation and suggestion phases could be a solution to avoid suggesting only the optimal paths: route perturbation was beneficial, resulting in shorter travel times and lower emissions in the city.

The study acknowledged that the situation in the real world is more complex, with multiple navigation apps coexisting simultaneously, each with its heuristics and representation of urban reality. The evidence suggests a need for algorithms that can exploit social and collective dimensions while simultaneously meeting individual needs. This challenge requires shifting from an individual to a collaborative and social paradigm, where the choices of non-rational or AI-assisted agents who exploit the system and their impact on the whole society are considered.

4. Conclusion

Urban trafic is a complex system where individual satis

faction is intimately bound to collective happiness, e.g., trafic is smooth for an individual if it is so for everyone. A system where individual interests go hand in hand with collective happiness. There is plenty of room for a better understanding of the impact of individual routing choices on the urban environment, as well as for studying how to design platform architectures and routing recommendations that influence citizens’ behaviour towards better aggregated outcomes. The challenge is to turn collective goals into an optimization target, whereby users transparently understand and accept recommendations, which can be individually sub-optimal but produce a more eficient collective outcome. Using (social) norms (such as minimising CO2 emissions) as targets for collective goals without using them as global constraints might be a way to achieve this goal.

Acknowledgments

This work has been partially supported by EU project

H2020 Humane-AI-net G.A. 952026, EU project H2020 SoBigData++ G.A. 871042, and by the CHIST-ERA grant CHIST-ERA-19-XAI-010, by MUR (grant No. not yet available), FWF(grant No. I 5205), EPSRC (grant No.

EP/V055712/1), NCN (grantNo. 2020/02/Y/ST6/00064), ETAg (grant No. SLTAT21096), BNSF(grant No. KΠ-06AOO2/5). ing trafic emissions during interruption and con- from personal travel in the uk, Energy Policy gestion, Transportation Research Part D: Trans- 36 (2008) 224–238. doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . port and Environment 43 (2016) 59–70. doi:h t t p s : e n p o l . 2 0 0 7 . 0 8 . 0 1 6 .

/ / d o i . o r g / 1 0 . 1 0 1 6 / j . t r d . 2 0 1 5 . 1 2 . 0 0 6 . [22] Y. Huang, N. Surawski, Y. S. Yam, C. Lee, J. Zhou, [13] F. Zheng, J. Li, H. van Zuylen, C. Lu, Influence B. Organ, E. Chan, Re-evaluating efectiveness of of driver characteristics on emissions and fuel vehicle emission control programs targeting highconsumption, Transportation Research Procedia emitters, Nature Sustainability 3 (2020). doi:1 0 . 27 (2017) 624–631. doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . 1 0 3 8 / s 4 1 8 9 3 - 0 2 0 - 0 5 7 3 - y . t r p r o . 2 0 1 7 . 1 2 . 1 4 2 , 20th EURO Working Group on [23] M. Böhm, M. Nanni, L. Pappalardo, Gross polluters Transportation Meeting, EWGT 2017, 4-6 Septem- and vehicle emissions reduction, Nature Sustainber 2017, Budapest, Hungary. ability (2022). doi:1 0 . 1 0 3 8 / s 4 1 8 9 3 - 0 2 2 - 0 0 9 0 3 - x . [14] L. Pappalardo, S. Rinzivillo, Z. Qu, D. Pedreschi, [24] G. Boeing, Osmnx: New methods for acquirF. Giannotti, Understanding the patterns of car ing, constructing, analyzing, and visualizing comtravel, European Physical Journal: Special Topics plex street networks, Computers, Environment 215 (2013). doi:1 0 . 1 1 4 0 / e p j s t / e 2 0 1 3 - 0 1 7 1 5 - 5 . and Urban Systems 65 (2017) 126 – 139. doi:h t t p s : [15] M. Nyhan, S. Sobolevsky, C. Kang, P. Robinson, / / d o i . o r g / 1 0 . 1 0 1 6 / j . c o m p e n v u r b s y s . 2 0 1 7 . 0 5 . 0 0 4 .

A. Corti, M. Szell, D. Streets, Z. Lu, R. Britter, [25] L. Vyas, N. Butakhieo, The impact of working from S. R. Barrett, C. Ratti, Predicting vehicular emis- home during covid-19 on work and life domains: sions in high spatial resolution using pervasively an exploratory study on hong kong, Policy Design measured transportation data and microscopic and Practice (2020) 1–18. emissions model, Atmospheric Environment 140 [26] L. Nagel, The influence of the covid-19 pandemic (2016) 352 – 363. doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . on the digital transformation of work, International a t m o s e n v . 2 0 1 6 . 0 6 . 0 1 8 . Journal of Sociology and Social Policy (2020). [16] J. Liu, K. Han, X. M. Chen, G. P. Ong, Spatial- [27] E. Ericsson, H. Larsson, K. Brundell-Freij, Optitemporal inference of urban trafic emissions based mizing route choice for lowest fuel consumption – on taxi trajectories and multi-source urban data, potential efects of a new driver support tool, TransTransportation Research Part C: Emerging Tech- portation Research Part C: Emerging Technologies nologies 106 (2019) 145 – 165. doi:h t t p s : / / d o i . o r g / 14 (2006) 369–383. doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . 1 0 . 1 0 1 6 / j . t r c . 2 0 1 9 . 0 7 . 0 0 5 . t r c . 2 0 0 6 . 1 0 . 0 0 1 . [17] C. K. Gately, L. R. Hutyra, S. Peterson, I. Sue [28] Z. Samaras, L. Ntziachristos, S. Tofolo, G. Magra, Wing, Urban emissions hotspots: Quantifying A. Garcia-Castro, C. Valdes, C. Vock, W. Maier, vehicle congestion and air pollution using mo- Quantification of the efect of its on co2 emisbile phone gps data, Environmental Pollution sions from road transportation, Transportation 229 (2017) 496–504. doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / Research Procedia 14 (2016) 3139–3148. doi:h t t p s : j . e n v p o l . 2 0 1 7 . 0 5 . 0 9 1 . / / d o i . o r g / 1 0 . 1 0 1 6 / j . t r p r o . 2 0 1 6 . 0 5 . 2 5 4 . [18] J. Chen, W. Li, H. Zhang, W. Jiang, W. Li, Y. Sui, [29] N. Arora, T. Cabannes, S. G. Subramaniam, Y. Li, X. Song, R. Shibasaki, Mining urban sustainable per- P. McAfee, M. Nunkesser, C. Osorio, A. Tomkins, formance: Gps data-based spatio-temporal analysis I. Tsogsuren, Quantifying the sustainability impact on on-road braking emission, Journal of Cleaner of google maps: A case study of salt lake city, 2021. Production 270 (2020) 122489. doi:h t t p s : / / d o i . o r g / [30] F. Perez-Prada, A. Monzón, C. Valdés, Managing 1 0 . 1 0 1 6 / j . j c l e p r o . 2 0 2 0 . 1 2 2 4 8 9 . trafic flows for cleaner cities: The role of green [19] Y. Sui, H. Zhang, X. Song, F. Shao, X. Yu, navigation systems, Energies 10 (2017) 791. doi:1 0 .

R. Shibasaki, R. Sun, M. Yuan, C. Wang, S. Li, Y. Li, 3 3 9 0 / e n 1 0 0 6 0 7 9 1 .

Gps data in urban online ride-hailing: A compar- [31] G. Cornacchia, M. Böhm, G. Mauro, M. Nanni, D. Peative analysis on fuel consumption and emissions, dreschi, L. Pappalardo, How routing strategies imJournal of Cleaner Production 227 (2019) 495 – pact urban emissions, in: Proceedings of the 30th In505. doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . j c l e p r o . 2 0 1 9 . ternational Conference on Advances in Geographic 0 4 . 1 5 9 . Information Systems, SIGSPATIAL ’22, Association [20] P. Guenther, G. Bishop, J. Peterson, D. Stedman, for Computing Machinery, New York, NY, USA, Emissions from 200 000 vehicles: a remote sens- 2022. doi:1 0 . 1 1 4 5 / 3 5 5 7 9 1 5 . 3 5 6 0 9 7 7 . ing study, Science of The Total Environment 146- [32] P. A. Lopez, E. Wiessner, M. Behrisch, L. Bieker147 (1994) 297 – 302. doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / Walz, J. Erdmann, Y.-P. Flotterod, R. Hilbrich, 0 0 4 8 - 9 6 9 7 ( 9 4 ) 9 0 2 4 9 - 6 . L. Lucken, J. Rummel, P. Wagner, Microscopic traf[21] C. Brand, B. Boardman, Taming of the few — the ifc simulation using SUMO, 2018. unequal distribution of greenhouse gas emissions

References

[1] Intergovernmental Panel on Climate Change , Climate Change 2014 : Mitigation of Climate Change: Working Group III Contribution to the IPCC Fifth Assessment Report , Cambridge University Press, 2015 , p. 599 - 670 . doi:1 0 . 1 0 1 7 / C B O 9 7 8 1 1 0 7 4 1 5 4 1 6 .

[2] H. Ritchie , Sector by sector: where do global greenhouse gas emissions come from ?, 2020 . Available at https://ourworldindata.org / ghg-emissions-by-sector.

[3] United Nations General Assembly , Transforming our world: the 2030 Agenda for Sustainable Development , Technical Report , 2015 . Accessed: 2021 -02- 23.

[4] J. Macfarlane , Your navigation app is making trafic unmanageable , IEEE Spectrum ( 2019 ) 22 - 27 .

[5] S. Siuhi , J. Mwakalonge , Opportunities and challenges of smart mobile applications in transportation , Journal of Trafic and Transportation Engineering 3 ( 2016 ) 582 - 592 .

[6] L. W. Foderaro , Navigation apps are turning quiet neighborhoods into trafic nightmares , The New York Times ( 2017 ).

[7] P. deSouza , A. Anjomshoaa , F. Duarte , R. Kahn , P. Kumar , C. Ratti , Air quality monitoring using mobile low-cost sensors mounted on trashtrucks: Methods development and lessons learned , Sustainable Cities and Society 60 ( 2020 ) 102239 . doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . s c s . 2 0 2 0 . 1 0 2 2 3 9 .

[8] H. S. Chong , S. Kwon , Y. Lim , J. Lee , Real-world fuel consumption, gaseous pollutants, and co2 emission of light-duty diesel vehicles , Sustainable Cities and Society 53 ( 2020 ) 101925 . doi:h t t p s : / / d o i . o r g / 1 0 .

1 0 1 6 / j . s c s . 2 0 1 9 . 1 0 1 9 2 5 .

[9] J. M. Luján , V. Bermúdez , V. Dolz , J. MonsalveSerrano , An assessment of the real-world driving gaseous emissions from a euro 6 light-duty diesel vehicle using a portable emissions measurement system (pems ), Atmospheric Environment 174 ( 2018 ) 112 - 121 . doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . a t m o s e n v . 2 0 1 7 . 1 1 . 0 5 6 .

[10] T. Chatterton , J. Barnes , R. E. Wilson, J. Anable , S. Cairns , Use of a novel dataset to explore spatial and social variations in car type, size, usage and emissions , Transportation Research Part D: Transport and Environment 39 ( 2015 ) 151 - 164 . doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . t r d . 2 0 1 5 . 0 6 . 0 0 3 .

[11] S. R. Kancharla , G. Ramadurai, Incorporating driving cycle based fuel consumption estimation in green vehicle routing problems , Sustainable Cities and Society 40 ( 2018 ) 214 - 221 . doi:h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / j . s c s . 2 0 1 8 . 0 4 . 0 1 6 .

[12] A. Choudhary , S. Gokhale , Urban real-world driv-