CC BY
92
Section 1. Innovation management
https://doi.org/10.29013/EJEMS-20-1-3-10
Jialu Hu,
11th Grade, Maryknoll High School, Honolulu, Hawaii E-mail: hujialu2003@gmail.com
EXPLORING THE USE OF ARTIFICIAL INTELLIGENCE TO SOLVE TRAFFIC CONGESTION
Abstract. This paper reviews the applications of artificial intelligence (AI) innovations in the field ofworldwide transportation management to explore the feasibility of using artificial intelligence to solve traffic congestion. Regional experimental applications of AI powered systems, applied to transportation planning, navigation, law enforcement, smart traffic lights, and driverless cars, provide promising results that AI can solve traffic congestion in the near future.
Keywords: Artificial Intelligence, Traffic Congestion, Smart Traffic Light, Induced Demand, Autonomous Car, AI Robotic Polices.
I. Introduction
As technology advanced over time, automobiles replaced the bike and became the primary transportation vehicle in the world, including China. For example, in 2018 alone, the number of cars sold reached around 86 million in 54 markets worldwide (Bekker [2]). The convenience of driving cars, however, brings with it the problem of traffic congestion. Based on a report by INRIX, in 2018, every American driver spent an average of 97 hours per year in traffic jams (Bradford [4]). According to INRIX, economically, traffic congestion cost the US $87 billion in lost productivity, an average of $1348 per driver, in 2018 (Le Beau [14]), and environmentally, traffic congestion wastes 19 gallons of fuel per commuter, which seriously depletes our valuable non-renewable resources ("Fact No. 897" [11]). As early as 2014, Peking University's National Development Research Institute assessed that traffic congestion costs Beijing $11.3 billion a year ("Traffic Jams Cost Beijing
$11.3b a Year" [20]). If we can solve the problem of congestion, people could spend their time in other meaningful pursuits rather than idly waiting in traffic, economic losses would be dramatically reduced, and natural resources could be saved for better use. How can we solve this problematic situation? One approach is that the problem caused by advanced technology can only be solved by more advanced technology in the form of artificial intelligence.
The exploration of artificial intelligence began in the mid-19th century. The word "Artificial intelligence" was coined by John McCarthy in a conference he organized. The Dartmouth Summer Research Project on Artificial Intelligence in 1956. Allen Newell, Cliff Shaw, and Herbert A. Simon presented the Logic Theorist, which was designed to mimic the problem-solving skills of a human and considered by many people as the first AI program, in the conference (Anyoha [1]). Typically, Artificial intelligence is classified into three groups: Artificial
Narrow Intelligence (ANI), Artificial General Intelligence (AGI), and Artificial Super Intelligence. (ASI) ANI is professional in one field, but only in the specific field that it is designed for. Every kind of brainwork performed by a human can be performed perfectly by AGI; they hold the ability of reasoning, problem-solving, are self-conscious, self-aware, and can independently solve problems and find solutions. According to Nick Bostrom, a philosopher at the University of Oxford and professor of artificial superintelligence, Artificial Super Intelligence "is much smarter than the best human brain in practically every field, including scientific creativity, general wisdom, and social skills" (Bostrom [3, P. 11]). Scientists have been innovating artificial intelligence because of its advantages and benefits. Artificial intelligence beats humans in speed, has unlimited storage space, and long term durability, not to mention its ability to upgrade, save, and edit (Urban [18]). In fact, a microprocessor's speed is 10 million times faster than human neurons. (2GHz V.S. 200Hz) Currently, our society is in the stage of extensive use of Artificial Narrow Intelligence as it increasingly becomes an indispensable part of our life. Many websites and navigation maps like Google, Baidu, Google Maps, Amap, and Baidu Maps have AIs in their system, while sports, finance, manufacturing, and even medical fields all use products with AI technologies. The question then is, can we employ artificial intelligence into the field of traffic control and use AI to help solve the problem of traffic congestion? This paper reviews the current application ofAI in traffic control and explores the feasibility of wider implementation.
II. Previous solutions to traffic congestions
In the past, governments usually relied on widening roads or expanding transportation systems to solve traffic congestion. However, as the induced traffic theory shows, increased roadway traffic capacity will also increase the number of cars on the road as the new road now allows more cars to use it simultaneously. Also called latent demand, this
theory is consistent with the economic concept of supply and demand. As the supply (expansion of current roadway) increases, the demand for using car as a means of primary transportation also increases. This theory was recognized as early as 1974 in New York by Robert Caro. His book, The Power Broker, he illustrated that roads actually aggravate congestion without a properly balanced transportation system, using evidence from the highway building programs of Robert Moses (Caro [5]). The reasoning behind induced demand is also explained byJ. J. Leeming in his book Road Accident: Prevent or Punish?, "Motorways and bypasses generate traffic, that is, produce extra traffic, partly by inducing people to travel who would not otherwise have done so by making the new route more convenient than the old, partly by people who go out of their direct route to enjoy the greater convenience of the new road, and partly by people who use the towns bypassed because they are more convenient for shopping and visits when through traffic has been removed" (Lemming [15]).
The case of the Katy Freeway in Texas is a good modern example, as it is the widest freeway in the world. The Katy Freeway underwent an expansion project at a cost of $2.8 billion between 2008 and 2011 intended to relieve traffic. However, after expansion, it was found that the travel times increased by 30 percent in the morning and 55 percent in the evening (Schneider [16]). This result was also consistent with research results from economist Matthew Turner of the University of Toronto and Gilles Duranton of the University of Pennsylvania. They compared the number of new roads and highways with the amount of traffic between 1980 and 2000 and discovered that when the road capacity increased a certain percent, the vehicle-kilometer traveled also correspondingly increased the almost identical percent. Turner and Duranton concluded that, "increased provision of roads or public transit is unlikely to relieve congestion" (Turner and Duranton [10]).
One more approach is to charge tolls to ease traffic during peak hours. For example, in Manhattan,
New York, drivers were required to pay a surcharge between 6 a.m. and 8 p.m. when they entered Manhattan south of 60th street and north of Battery Park or in the central business district. However, equity problem arose with this policy as most people driving into Manhattan were middle class and not the wealthy residents of Manhattan. In Long Island and Brooklyn, many residents were against this policy because they preferred driving because the subway stations were too far. This congestion relief pricing was also explored in China as early as 2016. China Youth Daily took a survey of more than 2000 participants, asking their opinion on a congestion relief fee. 68% of the participants believed that a congestion fee would only "cure the symptom, but not the disease" (J. Chen [7]).
Another approach to control the volume of traffic was implemented in Beijing, China, where cars with odd numbered license plates drive on odd days and even numbers on even days. This policy effectively reduces the amount of cars driven on the road in the short run, but will lose its effectiveness over time as those with enough financial resources could simply buy another car. Therefore this law could actually increase the amount of cars in the city. It is necessary to explore the root causes of traffic congestion to design more effective solutions to traffic congestion.
III. Causes of traffic congestion
I classified the causes of traffic congestion into three categories: road-related, human-related, and technology-related. Road-related congestion includes both the number of vehicles exceeding the capacity of the road and the poor design of the roads themselves. This type of traffic congestion usually happens in the period between 6a.m.- 9a.m., and 5p.m.- 7p.m., when the majority of commuters drive to work. This type of congestion mostly happens on vital communication lines, commercial central districts, roads with lots of traffic lights or works, and scenic spots. These places are where most drivers are traveling, and as numbers of cars converge together traffic congestion inevitably oc-
curs. Road-related congestion is primarily caused by people choosing vehicles over public transit. People prefer driving over public transit not only because of its privacy and comfort, but also for its convenience and flexible timing (Downs [9]). Another cause of road-related traffic jams is inefficient road designs. For example, in Beijing, the roads are designed as a radial from a central point, which was designed to be convenient for transportation between the suburbs and city center. However, during certain hours, all the vehicles from the surrounding suburbs are moving towards the city center, and the main roads in the city are filled with the inflow of traffic (C. Chen [6]).
Human-related traffic congestion includes all congestion resulting from human activities, whether consciously or unconsciously. It includes traffic congestions that could be avoided, such as accidents, which are unintentional, and traffic violations, which are mostly intentional. It's common to see aggregated traffic in both directions if an accident happens on one side of the highway. Similarly, when cars or pedestrians are stopped by the police to check their violations, the traffic is slowed down which leads to congestions. If a car is parked in a space that isn't designed for parking, all the cars passing will experience driving difficulties and slowed traffic.
Technology-related traffic congestion refers to any traffic jams that can be improved by more advanced technology, including utilizing the unprocessed traffic information to improve the inflexibility of traffic lights. Compared to road-related congestion, Artificial Intelligence can solve the problem of human-related and technology-related congestion in a significantly shorter period. In fact, AI is solving these two types of traffic congestion right now.
IV. Using ai to solve traffic congestions
While it is obvious that the use ofAI can directly help to reduce technology-related congestion, it can also help to alleviate congestion problems arising from the other two causes. Even though AI cannot help to physically change the design of existing roads,
it can be helpful in terms of guiding traffic volume over the road system, thus reducing congestion of the road-related and human-related types.
The foremost way AI can help road-related congestion is to promote public transportation and therefore reduce the number of vehicles on roads. In China, people's preferences for personal vehicles rather than buses is partially because of the bus system's unpredictability. Although dispatch schedules exist, many unexpected conditions occur; sometimes buses are not on time or there are too many people to fit on the bus. Anxious commuters may wait for a while then decide to take a taxi, which adds to the number of vehicles on the road. In China, Hisense developed the only autonomous dispatch system which combined artificial intelligence and big data analytics. It enhances the driving scheme of automatic generation and optimization, as well as automatic processing of unexpected events. The system greatly improves dispatch plans and reduces dispatch errors, which leads to more accurate bus scheduling. After the operation of the system in Chengdu, the efficiency of dispatching personnel increased by 200%, average passenger wait time was reduced from 9 minutes to 5 minutes, the complaint rate of passengers dropped by 20% and the public transportation consumer satisfaction rate of Chengdu also reached as high as 85%, the highest among all industry in the city (Q. Wang [21]). The system acts as a strong catalyst for people to give up personal vehicles and taxis and turn to public transportation, which will ease congestion by effectively reducing the amount of cars on the road.
Artificial Intelligence can solve human-related traffic congestion in two wars, 1) enabling driver-less cars and 2) using AI-powered law enforcing robots. The physical limitations of the human body contribute to many car accidents. Drunk driving, drowsy driving, weather conditions, or simply distractions from phones, drinks, or food can lead to accidents. It only takes a few seconds of distraction for car accidents to occur but usually requires at least
10-30 minutes to settle the accidents (J. Zhu [24]). When accidents occur, the damaged vehicles, being unable to move away, directly affect the traffic flow of one or more lanes and cause traffic congestion. The condition is worse when an ambulance and police cars are involved, especially during peak commuting hours. Driverless cars could directly solve these problems due to the characteristics of the machines. Driverless cars rely on artificial intelligence to integrate visual computing, radar, monitoring devices, and global positioning systems to operate vehicles safely and automatically, without any human initiative. In contrast to humans, artificial intelligence will never drive drunk, fatigued, distracted, or over the speed limit. This becomes especially critical when we examine the data. In 2018, 10,511 in 36,560 total traffic fatalities were alcohol-impaired fatalities according to U. S. Department of Transportation, National Highway Traffic Safety Administration ("2018 Fatal Motor Vehicle Crashes" [12]). If 90 percent of cars on U.S. roads become driverless, the number of accidents would fall from 6 million to 1.3 million and the death toll from 33.000 to 11.300 (Wibberley [23]).
With collaboration and investments from companies worldwide, the AI car is steadily developing. For example, Baidu's Apollo had 300 driverless cars on the roads with a total of 1.2 million miles driven by July 2019 (Sivalingam [17]). As a leading automotive company, Audi already unveiled its A8 autonomous driving feature, which not only allows drivers to take their hands off the steering wheel up to 37 mph but also was adapted in 2019 with more advanced object recognition and adaptive cruise control (Davies [8]). It will take time for autonomous cars to become commercially available to consumers, but AI cars are surely the future. Current issues, such as safety concerns, will be steadily overcome as AI technology continues to develop. On the other hand, consumer acceptance ofAI cars is very high, at least in China. In a survey I conducted in China with a sample size of534.86% of the respondents indicated interest in trying AI cars if available on the market.
Figure 1. Distribution of Willingness to Try AI Car If Available in Market
Successfully enforcing traffic rules will also ease traffic violations associated with pedestrians and vehicles, therefore resulting in better traffic flow. Pedestrians often run red light and cars often park illegally; these two activities are especially common in big cities where there are too many cars and walkers for police to control, inhibit, or punish. To address this problem, multiple cities in China started to employ AI powered systems. For example, Hangzhou (Zhejiang, China), implemented the AI product, Tianyao, in April 2018. Through automatic tracking identification and intelligent perception technology, Tianyao is equipped with real-time analysis capability similar to city-level large-scale cameras. It holds complete traffic event identification with the ability to discern reverse driving, non-motor vehicle traffic in motor vehicle lanes, pedestrian traffic in motor vehicle lanes, and other traffic anomalies within 20 seconds with 95% accuracy. In 2018, Tianyao controlled a total of249 monitoring ball machines in the pilot area, which covers nearly 700 sections of roads (the same area would require at least 200 police officers) and monitors the area 24 hours a day, 365 days a year, which cannot be achieved by manual patrols (Y. Zhu [25]). A similar system has been employed by the city of Shenzhen (Guangdong, China), but with more focus on pedestrians. In April 2017, a face recognition system was put into use on a trial road intersection to monitor pedestrians crossing at red
lights. This system was proven effective as during the six months of its implementation, 13.930 cases of pedestrian violations were caught and the number of violations decreased from about 150 cases per hour to 8 per hour ("The AI Traffic Police," [19]).
The city of Handan (Hebei, China) employed Robotic Police, which is the newest and most comprehensive AI system used in China. The Robotic Police integrated technology in AI, cloud computing, and multi-sensor fusion to achieve fully autonomous intelligence operations, regardless of weather and or time of day, to assist traffic police and help public services. The Robotic Police has three main types, Road Patrol Robot, Manage & Consult Robot, and Accident Alert Robot, each with a distinctive appearance. They can broadcast reminders, discern and report license plates during the patrol and when capturing violations, identify human faces, alert and expel illegal parkers, and answer questions. A single Robot Traffic Police can work 16 hours a day, and replace 1-2 auxiliary police personnel. It is estimated that one robot can save up to 150.000 yuan (about $21.429) a year (Wang and Chang [21; 22]) The use ofAI technology is very promising at alleviating human-related traffic congestion problems.
In the field of technology-related congestion, artificial intelligence is contributing significantly now and will into future. As early as 2017, Baidu map, Google maps and other map apps employed
AI technology. AI powered GPS systems can provide more accurate information about the current road situations and provide better routing for drivers to reduce traffic jams. Among the 534 respondents I surveyed, 68.35% acknowledge maps apps used AI technology, and as many as 98.51% consider the path planning function useful. For those who indicated that function as useful, 68.54% think the app automatically planned the quickest and shortest route after entering one or more destinations, 60.84% think it shows traffic jams areas and provides an opportunity to change the route when needed, 46.07% think it alerts them to road cameras, speed limits, time to turn a corner, and other notifications, only 3.93% choose none of the above.
Another use of AI could be extended to enable smart traffic lights. Most traffic lights in the world have a fixed amount of time to pass for pedestrians and drivers. Since the timing is consistent from the day it is set, traffic lights aren't flexible to the number of cars and pedestrians who are waiting to pass. On the other hand, the smart traffic light with AI would be able to allocate more time to the direction with heavier traffic flow at any given time. This technology is being piloted in areas of the world now. In America, Scalable Urban Traffic Control designed by Stephen Smith and other researchers of the Robotics Institute, Carnegie Mellon University enabled real-time optimization of traffic flow. It analyzes data collected from cameras and sensors to come up with the best phase change and controls the lights accordingly. When tested in nine intersections in the East Liberty region of Pittsburgh in June 2012, and contrasted with the previous condition in March 2012, the results were impressive. Average waiting time at intersections was reduced by over 40%, journey time decreased by 25%, average vehicle speed increased by 34%, number of stops
decreased by over 31%, and emission was reduced by 21% (Smith [18]). Because of these results, in the past three years, the program has expanded to 47 intersections in Pittsburg, PA. In China, two cities employed smart traffic light technology and realized the same successful result. The biggest taxi company in China, Didi, cooperated with the traffic police of Jinan (Shandong, China) to employ smart traffic lights in six intersections on Jingshi road which is the longest arterial street in China. Within one month, the work day average delay time decreased by 10.73% during morning peak hours, and 10.94% during evening peak hours. The number of stops also decreased, those in evening peak hours of the work day decreased by 8.7%, and the number of stops in the morning peak decreased by 6.7% (Zhu [24; 25]). In July 2017, Guangzhou traffic police piloted the "Internet + signal lamp" control and optimization platform, and directed the traffic in Guangzhou with the help of the traffic platform Austro-Tech. Before the pilot in the Haizhu district, there was a serious imbalance at the intersection between Nanhua Middle Road and Baogang Avenue. After implementing the system, the congestion was reduced by 25.75% from 9 a.m.- 1 p.m. and by 11.83% from 3 p.m.- 8 p.m. (Qu, 2018).
V. Conclusion
With efficiency, consistency, endurance, and accuracy, artificial intelligence can help solve traffic congestion. Innovation from both the U.S. and China shows that AI is capable of solving traffic congestion through five different approaches: planning, navigation, law enforcement, smart traffic lights, and autonomous vehicles. AI may be the only approach to solve traffic congestions with no side effects and it is very likely that the current initiatives will move from regional experimental tests to extensive national implementation.
References:
1. Anyoha Rockwell. "Can machines Think?" sitn.hms.harvard.edu, cited 18 Jul 2019. URL: http://sitn. hms.harvard.edu/flash/2017/history-artificial-intelligence
2. Bekker Henk. "2018 (Full Year) International: Worldwide Global Market Analysis." Best-selling-cars. com, 23 Feb, 2019. - cited 22. - Jul, 2019. URL: https://www.best-selling-cars.com/global/2018-full-year-international-worldwide-car-sales-and-global-market-analysis
3. Bostrom Nick. "How Long Before Superintelligence?" Linguistic and Philosophical Investigations, 2006.- Vol. 5.- No. 1.- P. 11-30.
4. Bradford Ben. "The Average American Spend the Equivalent of 2.5 Work Weeks in Traffic." Marketplace. org, 13 Feb, 2019.- cited 21.-Jul, 2019. URL: https://www.marketplace.org/2019/02/13/report-aver-age-american-spends-equivalent-25-work-weeks-traffic/
5. Caro Robert. The Power Broker: Robert Moses and the Fall of New Tork.- New York: Knopt. 1974.
6. Chen Chen. "The Roads get wider, Beijing get more jam. City Planning & City Development." Wenku. baidu.com, 13 Aug, 2018.- cited 08.- Aug, 2019. URL: https://wenku.baidu.com/view/d853c4624a73f-242336c1eb91a37f111f1850d28.html
7. Chen Junsong. "Charging congestion fees to treat congestion: nearly 70 percent of people think that cure the symptom but not the disease." Xinhuanet.com, 02 Feb, 2016.- cited 16.- Nov, 2019. URL: http://www.xinhuanet.com//politics/2016-02/02/c_128693685.htm
8. Davies Alex. "Americans Cant' Have Audi's Super Capable Self-Driving System." Wired.com, 15 May,
2018.- cited 02.- Dec, 2019. URL: https://www.wired.com/story/audi-self-driving-traffic-jam-pi-lot-a8-2019-availablility
9. Downs Anthony. Srill Stuck in Traffic: Coping with Peak-Hour Traffic Congestion. Washington D. C.: The Brookings Institution. 2004.- 43 p.
10. Duranton Gilles, and Matthew A. Turner. The Fundamental Law of Road Congestion: Evidence from US Cities. American Economic Review,- 101 (6). 2011.- P. 2616-52.
11. "Fact No. 897.- November 2. 2015. Fuel Wasted in Traffic Congestion." Energy.gov.- 2 Nov, 2015.- cited 22.-Jul, 2019. URL: https://www.energy.gov/eere/vehicles/fact-897-november-2-2015-fuel-wasted-traffic-congestion
12. "2018 Fatal Motor Vehicle Crashes: Overview." Crashstats. nhtsa. dot.gov, Oct, 2018.- cited 22.- Jul,
2019. URL: https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812826
13. Ju Ming. "Artificial intelligence with traffic lights, significantly improve traffic." Sohu.com, 30 July, 2018.-cited 14.- Aug, 2019. URL: http://www.sohu.com/aZ244183810_505843
14. Le Beau Phil. "Traffic Jams Cost US $87 billion in Lost Productivity in 2018 and Boston and DC had the Nation's Worst." CNBC. Com, 12 Feb, 2019.- cited 21.- Jul, 2019. URL: https://www.cnbc. com/2019/02/11/americas-87-billion-traffic-jam-ranks-boston-and-dc-as-worst-in-us.html
15. Lemming J. J. and Mackay G. M. Road Accidents: Prevent or Punish? Cassell. 1969.
16. Schneider Benjamin. "Citylab University: Induced Demand." Citylab.com, 06 Sep, 2018.- cited 21.-Oct, 2019. URL: https://www.citylab.com/transportation/2018/09/citylab-university-induced-de-mand/569455
17. Sivalingam James. "What does the 1m miles milestone mean for Baidu's Autonomous Cars?" Techwireas-ia.com, 4Jul, 2019.- cited 02.- Dec 2019. URL: https://techwireasia.com/2019/07/what-does-the-1m-miles-milestone-mean-for-baidus-autonomous-cars
18. Smith S. F., et al. SURTRAC: Scalable Urban Traffic Control. 1, figshare, 01.- Jan, 2018. Doi:10.1184/ R1/6561035.v1.
19. "The AI traffic police caught 125 red light runs on their first day on duty. How can AI "end" public transport chaos" sohu.com, 05 May, 2018.- cited 07.- Aug 2019. URL: https://www.sohu. com/a/230560296_115035
20. "Traffic Jams Cost Beijing $11.3b a Year." Chinadaily.com.cn, 29 Sep, 2014.- cited 07.- Nov, 2019. URL: https://www.chinadaily.com.cn/china/2014-09/29/content_18679171.htm.
21. Wang Qiong. "Intelligence Provide Chengdu buses a Flying Wing." Wenku.baidu.com, Nov, 2010.- cited 20.- Aug 2019. URL: https://wenku.baidu.com/view/c992b2ea172ded630b1cb653.html.
22. Wang Wei and Chang Zheng. "National first! Robot traffic police in handan are on duty 24 hours duty." Bajihao.baidu.com, 07 Aug, 2019.- cited 08.- Sep, 2019. URL: http://baijiahao.baidu.com/s?id=1641 202947014142406&wfr=spider&for=pc
23. Wibberley Ryan. "Uber Shows That the Future is Driverless, But Why?" Forbes.com, 15 Nov, 2016.- cited 07.- Aug, 2019. URL: https://www.forbes.com/sites/ryanwibberley/2016/11/15/uber-is-showing-us-that-the-future-is-driverless-but-why/#227fc4205520
24. Zhu Yinling. "Hangzhou city brain takes 20 seconds to find road traffic incident." Baijiahao.com, 10 Apr, 2018.- cited 14.- Aug, 2019. URL: https://baijiahao.baidu.com/s?id=1597326901644369996&wfr= spider&for=pc.
25. Zhu Jie, et al. A New Framework of Vehicle-Road Collaboration for Zero Congestion. Presented at 2018. SAECCE.- 8 Nov, 2018. URL: https://wenku.baidu.com/view/f85b008302d8ce2f0066f5335a-8102d277a26102.html
