Problem SPACE
Optimizing traffic signal control is a problem that lies at the heart of the field of transportation systems engineering. Analyzing data from 494 urban areas in the U.S., the Texas Transportation Institute’s (TTI) newly-released 2019 Urban Mobility Report puts the nationwide cost of traffic congestion at $166 billion of time and fuel wasted per year. Each individual commuter, on average, loses 54 hours per year and spends over $1000 on wasted fuel due to traffic delays. Amidst urgent calls for action to avert the worst impacts of climate change, there is clearly a need for innovative approaches to developing more efficient traffic control systems.
Enter Arts Integration. The Green Light SONATA (Signal Operation with Neuro-fuzzy Acoustic Tuning Application) seeks to harness the practical decision-making skills exercised by musicians in the act of improvisation and apply this expertise to traffic congestion, sonically represented. Understanding musical improvisation as a social behavior, we begin with a hypothesis that the musical intuition exhibited by performers during collective improvisational performance could be usefully applied to a sonic representation of traffic congestion at an intersection (which can itself be considered, in some ways, a collective improvisational performance). This hypothesis spurs inquiry back into music as we explore which specific elements of music-making contribute to a successful musical experience, and how these elements might inform traffic control systems. By sonifying data from traffic simulation software, having musicians respond musically to the sonified traffic data, and collecting and comparing the musically-informed control data with that from traffic controlled by traditional algorithms, we explore the feasibility of utilizing musical behavioral data to improve traffic efficiency through arts-integrated operation.
CONTEXT
Virtually everyone has experienced waiting long minutes for a traffic light to turn green, even with no traffic from competing directions. Avoiding these unnecessary delays could save each of us time, money, and in some cases, our pleasant disposition in our next human encounter. However, the problem of minimizing delay through optimizing signal control presents several unique challenges. Intersection controllers are commonly programmed according to a version of Webster’s formula (first published in 1958) that calculates an “optimal” green light cycle length based on the proportion of traffic flow as well as the maximum capacity in all conflicting directions. In many instances, traffic cycle changes end up being based on timed phases and/or sensor data at a single intersection, without taking into account how traffic flow from one intersection will affect other intersections in a systematic way. Real-life traffic flow is subject to high complexity of variables, both predictable and unpredictable. In heavily congested areas, small problems can quickly grow to impact traffic at multiple intersections.
In recent years researchers have begun to explore using machine learning to better account for the complexities of real-life traffic flows. Using this approach, a computer is trained on large sets of data to accurately predict the outcome of complex situations. However, machine learning may not be the way to achieve the best possible outcome in this case (defined as the control that would minimize traffic delays). Machine learning heavily depends on the existence of training datasets that are already known to provide the optimal mapping between traffic conditions and the most appropriate control solutions. Since such datasets are hard to find, we suggest that an entirely different approach could offer an alternative model of “optimal” traffic flow that could also inform machine learning.
ARTS INTEGRATION
Our project aims to provide an arts-integrated approach to traffic control, first at a single intersection, and eventually at the network level. Working with an industry-standard traffic simulation software (VISSIM), we translate the incoming traffic at one intersection into musical sound (see accompanying media). Each of the eight incoming traffic streams is assigned a unique musical pitch, which sounds as a plucked tone as a vehicle approaches an intersection, and a sustained tone, increasing in volume, as cars wait at the intersection.
Sonifying data from traffic simulation software allows the human ear to discern patterns that may not be salient in visual and/or numerical data. We ask trained musicians to listen and respond musically to this sonified data - to control the traffic by playing the pitch assigned to a given traffic direction, giving those vehicles the “green light.” Over the course of repeated interaction with the system, musicians become faster and more strategic in minimizing traffic delays. Through this process, the strategies that musicians demonstrate while improvising in response to auditory cues offer alternative solutions to what has become the standard approach to traffic systems control. The musical strategies are translated back into traffic control in the simulator and measured against Webster’s optimal control.
To date, we have developed a software interface between commercial traffic simulation software and musical input devices, designed and conducted an experiment to test musicians’ response to sonified traffic data, and conducted statistical data analyses comparing performance data of the signal when operated by musical versus traditional control.
Results from our initial experiment are promising, with the musical control of two of the three musicians’ responses resulting in less total traffic delay than Webster’s optimal control. Additionally, we found that all musicians performed better when responding to auditory cues only, as compared to responding to both auditory and visual information from the traffic simulation software. This suggests that indeed, a musical approach could offer an innovative and potentially advantageous disruption to conventional methods of traffic systems control.
INNOVATION
Our initial project activities leave plenty of room to deepen multidisciplinary integration. While we have succeeded in “gamifying” traffic control in a musical way for a single traffic intersection, we have yet to develop and run an experiment applying this approach at a network level. When we initially conceived this project, we imagined that musical sound and traffic systems would converge in such a way that the most optimal traffic solution would also sound deeply satisfying in some way to the musicians involved. While there is beauty in the music resulting from our initial excursions, we feel compelled to interrogate further the types of sound that can be produced by data sonification and by human musical interactions within this framework.
In the future we plan to modify the sounds of different variables within the traffic sonification, and to sonify data from multiple intersections simultaneously, to add levels of complexity to the musical action and ultimately gather more robust data to inform a more efficient learning approach to traffic control. Forthcoming experiments with traffic data sonification will take advantage of the human ability to simultaneously perceive multiple auditory streams, including changes in pitch, rhythm, and timbre, and to exercise complex judgment in responding to these auditory streams in real time.
Larger questions raised by this project extend beyond the fields of musical performance and civil engineering, reaching into philosophical inquiry into aesthetics and utility, and how each aspect influences our perceptions of musical worth and beauty.
RESEARCH PROCESS
Our research process is unique in bringing together specialists from distinct disciplines and initiating conversations that necessitate deep reflection on the foundations of each discipline and how one might inform the other. Data sonification has gained significant traction across scientific disciplines as a way to enhance human understanding of complex information; however we are aware of no other instances in the literature of human interaction with data through artistic (musical) action to inform machine learning. As such, our work represents an innovative process that could be quite generative of ideas for researchers across disparate fields and disciplines.
Over the course of this project we as researchers have found ourselves translating concepts from one field to another, and mining the breadth of our team’s experience as researchers and performing musicians. With a team that includes civil engineers, a composer/performer and an ethnomusicologist, ideas from historical rules of Western music theory are entertained alongside artificial intelligence, multicultural musical practices, and industry-standard NEMA traffic control phasing. In the absence of a clear map forward, our team has followed intuition into productive discussion through which we collectively determine which ideas are immediately worth pursuing and which are tabled for possible future consideration. We hope this project will inspire others to undertake further work exploring transportation, sound, and society.
BIBLIOGRAPHY
- Abbas, M. and A. Sharma. “Configuration of Traffic-Responsive Plan Selection System Parameters and Thresholds: Robust Bayesian Approach.” Transportation Research Record: Journal of the Transportation Research Board 1867 (2004). 233–242.
- Abbas, M., N. Chaudhary, G. Pesti, and A. Sharma. “Configuration Methodology for Traffic-Responsive Plan Selection: A Global Perspective.” Transportation Research Record: Journal of the Transportation Research Board 1925 (2005). 195–204.
- Abbas, M., G. Pesti, N. A. Chaudhary, and P. Li. “Illustrative Field Configuration and Evaluation of Traffic-Responsive Control.” J. Transp. Eng. 135, no. 9 (2009). 591–599.
- Cai, C., C. K. Wong, and B. G. Heydecker. “Adaptive traffic signal control using approximate dynamic programming.” Transportation Research Part C: Emerging Technologies 17, no. 5 (2009). 456–474.
- Chou, C.-H. and J.-C. Teng, “A fuzzy logic controller for traffic junction signals,” Information Sciences 143, no. 1-4 (2002). 73–97.
- Gartner, N.H., L. Zhang, and H. Li, “Comparative evaluation of three adaptive control strategies: OPAC, TACOS, and FLC.” Transportation Research Board 85th Annual Meeting, 2006.
- Goudarzi, Visda. “Designing an Interactive Audio Interface for Climate Science.” IEEE MultiMedia 22, no. 1 (2015). 41-47.
- Gresham-Lancaster, Scot. “Relationships of Sonification Music and Sound Art”. AI & Society 27 (2012). 207-212.
- Grond, Florian and Hermann, Thomas. “Aesthetic Strategies in Sonification”. AI & Society 27 (2012): 214-222.
- Junchen, J. and M. Xiaoliang. “A Learning-based Adaptive Signal Control System with Function Approximation.” IFAC-PapersOnLine 49, no. 3 (2016). 5–10.
- Mirchandani, P. and K. L. Head. “A real-time traffic signal control system: Architecture, algorithms, and analysis.” Transportation Research Part C: Emerging Technologies 9, no. 6 (2001). 415–432.
- Nichols, C., and Lorang, M., "Sound of Rivers: Stone Drum: Translating limnology to performed and fixed multimedia." Proceedings of the Korean Electroacoustic Music Society Annual Conference, Seoul National University, Seoul, S. Korea, October 2014.
- Nichols, C., et al., “Sound of Rivers: Stone Drum: a Multimedia Collaboration, with Sonified Data, Computer-Processed Narration, and Electric Violin.” Proceedings of the International Computer Music Conference (ICMC), Athens, Greece, September 2014.
- Parker, Jennifer Andrea. Composing [De]Composition: Data Sonification for Sound Art and Music Composition. Ph.D. Dissertation. Riverside: University of California, Riverside, 2016.
- Pesti, G., M. M. Abbas, and N. Chaudhary. “Traffic State Classification in Condition-Responsive Traffic Control.” 11th International Conference on Intelligent Engineering Systems (2007). 33–37.
- Schrank, David, Bill Eisele and Tim Lomax. 2019 Urban Mobility Report. August 2019: Texas A&M Transportation Institute. Accessed August 25, 2019. Retrieved from https://static.tti.tamu.edu/tti.tamu.edu/documents/mobility-report-2019.pdf
- Sen, S. and K. L. Head. “Controlled Optimization of Phases at an Intersection.” Transportation Science 31, no. 1 (1997). 5–17.
- Srinivasan, D., M. C. Choy, and R. L. Cheu. “Neural Networks for Real-Time Traffic Signal Control.” IEEE Trans. Intell. Transport. Syst. 7, no. 3 (2006). 261–272.
- Zakariya, Ahmed Y., and Sherif I. Rabia. “Estimating the Minimum Delay Optimal Cycle Length Based on a Time-Dependent Delay Formula.” Alexandria Engineering Journal 55, no. 3 (2016): 2509–14.
- Zhu, F., H. A. Aziz, X. Qian, and S. V. Ukkusuri. “A junction-tree based learning algorithm to optimize network wide traffic control: A coordinated multi-agent framework.” Transportation Research Part C: Emerging Technologies 58 (2015). 487–501.