Download PDF Spring Bridge on Sustainable Smart Cities March 15, 2023 The world’s cities face increasing threats from natural disasters, aging infrastructure, traffic, and resource constraints. The articles in this issue examine smart infrastructure, sustainability, net zero carbon options, and autonomous driving, among other approaches to smart and sustainable cities. Smart Infrastructure for Autonomous Driving in Urban Areas Thursday, March 16, 2023 Author: Guyue Zhou, Guobin Shang, and Ya-Qin Zhang VICAD can enable safer, more comfortable, more energy-efficient, and more environmentally friendly driving in smart cities. Autonomous driving (AD) is recognized as a core technology to advance the transportation paradigm shift.[1] Studies have shown that, in the United States, AD may not only reduce up to 94 percent of traffic fatalities by eliminating accidents that are due to human error (NHTSA 2015) but also free up 50 minutes each day per driver (NHTSA 2020). It also has the potential to create a new $1.5 trillion industry by 2030 (Gao et al. 2016). Recent work in vehicle-infrastructure cooperative autonomous driving (VICAD) significantly augments the capability and effectiveness of AD through close coordination with pedestrians, other vehicles, roads and traffic, and the cloud (AIR and Baidu 2021). In this article we discuss VICAD’s advantages and challenges to help make autonomous driving a reality with large-scale economic deployment. Introduction The automotive and transportation industry will undergo a tectonic shift in the next decade with advances in connectivity, automation, sharing, and electrification. Autonomous driving presents a historic opportunity to transform the academic, technological, and industrial landscape through advanced sensing and actuation, high-definition mapping, new machine learning algorithms, and smart planning and control. The international Society of Automotive Engineering (SAE 2016) defines levels of autonomy from none (Level 0) to full (Level 5). While significant progress has been made in R&D to support autonomous driving, high-level (Level 4+) AD road testing and commercial trials, led by China and the United States, show that there remain technical challenges and policy issues to expand this capability on urban streets. Since the US Department of Transportation initiated the Vehicle Infrastructure Integration (VII) program in 2003 (renamed IntelliDrive in 2009) to improve safety, mobility, and convenience (US DOT 2010), other countries and regions—such as Japan (Strategic Headquarters 2013), China (State Council of PRC 2015), and Europe (ERTRAC 2022)—have made remarkable progress on VICAD deployment. As AD decreases its marginal revenue while solving long-tail problems, VICAD will be the most likely scenario in the future. AD mainly relies on vehicles’ on-board sensors, computing power, and drive-by-wire systems for environmental perception, intelligent decision making, and control. VICAD goes a step further, integrating smart vehicles with pedestrians’ IoT devices (e.g., smartphones, smartwatches), roadside sensors, cloud-based data and computing, and other connected equipment that provides effective information for autonomous operation. With a much broader array of spatial sensing sources for perception, access to temporal/historical information for decision making, and the capacity to coordinate multiple transportation participants, VICAD is capable of more reliable perception to make smarter decisions in real time and to enable collaborative operation among multimodal transportation participants. VICAD can enable safer, more comfortable, more energy-efficient, and more environmentally friendly driving, playing a significant role in the transportation system for modern smart cities. In this article we briefly describe the state of VICAD and explain how it improves on AD’s capacities in driving safety, operational design domain, and traffic efficiency. We point out challenges to VICAD’s continued progress and suggest next steps. VICAD Stages of Implementation In 2019 the government of China announced the development of VICAD (CHTS 2019). A subsequent white paper (AIR and Baidu 2021, published by Tsinghua University)[2] summarized VICAD’s development in three stages (figure 1), considering technical maturity and passenger understanding of collaborative functions. As AD decreases its marginal revenue while solving long-tail problems, VICAD will be the most likely scenario in the future. In the stage of collaborative information interaction (stage 1), on-board units (OBUs) communicate with roadside units (RSUs) to exchange information (e.g., traffic light status) between vehicles and roads with either dedicated short-range communications or cellular vehicle-to-everything. In the stage of collaborative perception (stage 2), with the rapid growth of roadside perception -capability, smart roads can be either a complementary source of information (e.g., for blind spots of on-board sensors) or a redundant source (e.g., for low-height obstacles). For different grades of maturity, collaborative perception is further classified as primary (stage 2-1) or advanced (stage 2-2). The latter is required to support Level 4 AD (i.e., VICAD) with enhanced coverage density, sensor diversity, inspection accuracy, and positioning precision. In stage 3, collaborative decision and control, the smart road can make some decisions and exercise limited control of moving vehicles (e.g., to ensure mandatory yielding to an ambulance) within a defined scope of authority. Before open roads qualify for collaborative decision and control (stage 3-2), a transitional stage (stage 3-1) authorizes designated smart roads with conditional implementation in AD-only (e.g., AD--exclusive lanes) or enclosed (e.g., parking lots) areas. For real-world deployment, the major development stage of VICAD is upgrading from stage 1 to 2-1. In the United States, industry leaders are taking steps to adapt to VICAD and have articulated roadway needs for AD vehicles. Trials have been conducted in California, Arizona, and other states (Toh et al. 2020), and the Michigan Department of Transportation (MDOT 2020) is working with Cavnue to develop a corridor with smart infrastructure to support driving automation between Detroit and Ann Arbor. In China several large cities have established VICAD test sites to facilitate R&D, policy formulation, and research (BICMI 2022). In 2020 the Beijing High-level Automated Driving Demonstration Area (BJHAD), built in Yizhuang, became the world’s first high-level VICAD-based demonstration area (figure 2). Within BJHAD’s 60 km2, there are 332 intersections fully covered by smart infrastructure and more than 300 high-level autonomous vehicles for taxi service and open road tests. Driving Safety According to statistics of the World Health Organization, around 1.3 million people lose their lives each year due to road traffic accidents.[3] A United Nations General Assembly resolution calls for halving the global number of traffic casualties by 2030.[4] Given AD’s algorithmic complexity, driving safety tops the research topic list in transportation (Toh et al. 2020) and is the most critical factor hindering AD’s large-scale deployment. To enhance AD safety, the International Organization for Standardization (ISO 2019) proposed a “safety of the intended functionality” (SOTIF) framework to reduce risks from both systemic and random hardware failures for AD vehicles. A SOTIF scenario presents environmental and traffic conditions, including how the AD vehicle responds (e.g., emergency braking ahead, traffic lights, a person or animal crossing the road). Scenarios are categorized as known safe, known unsafe, unknown safe, or unknown unsafe. With stage 2 VICAD, collaborative perception enhances AD in blind spots and sensor failure (e.g., camera-obstructed) conditions. This capacity transforms unknown into known SOTIF scenarios. Moreover, with stage 3 VICAD, collaborative decision making and control can be used to determine the right of way for AD in multivehicle interactions and unexpected road conditions (e.g., road construction, traffic accident). This too may transform unsafe scenarios into safe ones in SOTIF. For safety reasons, it is not realistic to conduct large-scale real-world experiments to evaluate the safety benefits of different AD strategies. A high-fidelity simulator is therefore essential to analyze SOTIF systematically and quantitatively. Table 1 presents the comparative experimental results of AD and VICAD based on a digital approximation of typical traffic scenarios in BJHAD (AIR and Baidu 2021). The results show that VICAD significantly improves driving safety in high-dynamic scenarios. Operational Design Domain The operational design domain (ODD) defines all conceivable individual and overlapping conditions, use cases, restrictions, and scenarios that an AD vehicle might encounter (US DOT 2016). A vehicle’s level of automation depends on not only the AD level but also the ODD in which the AD is capable of operating. VICAD can help expand the AD ODD with additional information acquired from other connected nodes (e.g., pedestrians, roads, cloud servers) and external control commands from authorized remote controllers (e.g., an emergency vehicle, temporary RSU guidance, cloud-based drivers). We offer an illustration of the effectiveness of VICAD implementation based on the work of Baidu, an enterprise with the largest number of AD test vehicles and the highest AD test mileage annually in Beijing (BICMI 2021). Figure 3 shows typical AD failure cases, which can be easily resolved when VICAD draws from online observations (i.e., based on networked sensors), offline knowledge, and even human intelligence. From the spatial perspective, VICAD can offer additional information as a vehicle’s second viewpoint. Despite rapid progress in AD perception, visibility can still be compromised or reduced in long-range or occluded areas (1st column, figure 3). VICAD can record and transmit historical traffic data, such as the successful strategies of other vehicles, to an autonomous vehicle whose path is blocked (2nd column, figure 3), thereby enhancing AV perception and decision making. In addition, VICAD can download driving--related data from cloud servers. Thanks to road anomaly information uploaded by either smart roads or human drivers’ reports (3rd column, figure 3), AD vehicles can be aware of dynamic traffic conditions in real time to adjust their routes automatically. Most importantly, VICAD enables the cloud-based driver, which is the requisite to expand ODD for high-level AD. The cloud-based driver provides vehicles with real-time driving assistance for “extreme” conditions (i.e., an unforeseen situation that the vehicle doesn’t know how to handle; 4th column, figure 3). VICAD employs an AI-based discriminator in the cloud server to monitor AD status. When the vehicle is not functioning well (e.g., because of an exceeded perception uncertainty threshold), the cloud server will initiate appropriate measures to provide immediate service and assistance. Traffic Efficiency Autonomous driving could improve traffic flow by up to 35 percent by coordinating AD vehicles to keep traffic moving smoothly (Hyldmar et al. 2019). In different development stages, VICAD can improve traffic efficiency by coordinating traffic lights, vehicles, and even travel demands. With stage 1 VICAD, a straightforward way to improve traffic efficiency is through control of traffic lights, an approach that has been widely and globally applied. In BJHAD, Baidu has set up a traffic light control system for 315 adjacent intersections, with 51 trunk road coordi-nators and 36 controller-deployed single-node adaptive traffic lights. As an optimized result of VICAD, travel time on main roads is decreased by 10.4 percent on average, and the queue length at single-node adaptive intersections is decreased by 19.6 percent on average. With stage 2 VICAD, collaborative perception among multiple adjacent intersections leads to a better prediction of traffic flow and improved traffic efficiency. With stage 3, collaborative decision and control can jointly control traffic lights and moving vehicles. Additional roadway properties (e.g., variable lanes) and travel demands (e.g., robo-taxi flow control) may further contribute to optimization. A more complex system means greater deployment difficulty, but also better collaborative control and traffic efficiency. Challenges and Next Steps VICAD has obvious advantages, but its real-world deployment faces great challenges: Time and cost to build large-scale roadside infrastructure, which varies in different countries and cities. Field tests are of irreplaceable value for VICAD valida-tion and verification, so more large-scale test zones (like BJHAD) are required. Difficulty in coordinating different innovation cycles for AD technology and city infrastructure. With increasing large-scale testing and trials, AD’s rapid development progress constantly upgrades technical assumptions for roadside infrastructure. Ethics and laws not ready for both AD and VICAD in many respects. Two typical concerns are data use for AI algorithms without sufficient attention to data privacy and individual/collaborative decision making to minimize inevitable traffic accidents (as captured in “the trolley problem”; Basl and Behrends 2019). Lack of a VICAD platform, including datasets, standards, and other resources. Unlike available public AD datasets and jobs, there are few educational and professional opportunities in VICAD. Based on public and proprietary statistics of AD intelligence, figure 4 compares the expected evolution of VICAD and AD (AIR and Baidu 2021). VICAD is expected to take the lead in launching high-level AD products, for its advantages in long-tail problem solving. Will VICAD help AD become more publicly acceptable? A positive feedback loop is needed: safer VICAD enables wider commercialization with more customers and more AD data. The feedback data also further improve AD performance. Communication and coordination among different communities will be essential (Harrington et al. 2018), engaging those involved in AD/VICAD technology, the automotive industry, roadway infrastructure, and government. The VICAD technology community should follow industry standards and design a progressive and flexible technical roadmap to fulfill both short-term return on investment and long-term compatibility of new technology in roadside infrastructure planning. The automotive communities should adjust designs based on infrastructure planning, allocate resources for VICAD R&D, build VICAD’s scientific research infrastructure (e.g., datasets, benchmarks, challenge), and attract talent to this new area. The government should lead and carry out or sponsor research, and is also responsible for formulation of policies, regulations, and standards, all of which are key to the development and implementation of VICAD. Research funds should be allocated to encourage long-term university and industry collaboration for VICAD. BJHAD is a positive government-led example to facilitate collaboration among different players and expedite VICAD innovation. Recent VICAD efforts in Yizhuang are pioneering worldwide R&D to -deliver project Apollo Air (AIR and Baidu 2021), which enables high-level AD with roadside sensing capability; a publicly accessible dataset[5] for VICAD (Yu et al. 2022); and an open-source operating system for RSUs.[6] Conclusion The development of high-level AD technology is facing great challenges even as massive long-tail problems are being solved. By connecting vehicles with roadway infrastructure, the cloud, and other smart devices, VICAD can help solve critical problems by improving AD safety, surmounting ODD limitations, and optimizing traffic efficiency, environmental perception, decision making, and control execution. There is strong evidence that VICAD can facilitate AD adoption and improve transportation in smart cities. Acknowledgments We are sincerely grateful for Bridge managing editor Cameron Fletcher’s in-depth review and skillful editing of this article. We also appreciate that the Baidu Apollo team and BJHAD provided active deployment support and shared valuable data for our VICAD research, and AIR’s team contributed expertise in DAIR-V2X, AIROS, and VICAD architecture. References AIR [Institute for AI Industry Research, Tsinghua University] and Baidu. 2021. Towards Autonomous Driving: Key Technologies and Prospects of Vehicle-Infrastructure Cooperated Autonomous Driving. Available at https://air.tsinghua.edu.cn/info/1056/1920.htm. Basl J, Behrends J. 2019. Why everyone has it wrong about the ethics of autonomous vehicles. The Bridge 49(4):42–47. BICMI [Beijing Innovation Center for Mobility Intelligent Co., Ltd.]. 2022. Beijing Autonomous Vehicle Road Test Report (2021). Available in Chinese at www.mzone.site/Uploads/Download/2022-02-10/6204917913b55. pdf. CHTS [China Highway and Transportation Society]. 2019. Connected Automated Vehicle Highway (CAVH): A Vision and Development Report for Large-Scale -Automated Driving System (ADS) Deployment. Working Committee of Automated Driving. ERTRAC [European Road Transport Research Advisory Council]. 2022. Connected, Cooperative, and Automated Mobility Roadmap. Brussels. Available at https://www.ertrac.org/wp-content/uploads/2022/07/ERTRAC- CAD-Roadmap-2019.pdf. Gao P, Kaas H-W, Mohr D, Wee D. 2016. Automotive -revolution–perspective towards 2030: How the convergence of disruptive technology-driven trends could transform the auto industry. McKinsey & Company. Harrington RJ, Senatore C, Scanlon JM, Yee RM. 2018. The role of infrastructure in an automated vehicle future. The Bridge 48(2):48–55. Hyldmar N, He YJ, Prorok A. 2019. A fleet of miniature cars for experiments in cooperative driving. Internatl Conf on Robotics & Automation, May 20–24, Montreal. ISO [International Organization for Standardization]. 2019. Road vehicles—Safety of the intended functionality (ISO Standard No. 21448:2019). Online at https://www.iso.org/standard/70939.html. MDOT [Michigan Department of Transportation]. 2020. CAV corridor. Lansing. Online at https://www.michigan.gov/mdot/travel/mobility/initiatives/ cav-corridor. NHTSA [National Highway Traffic Safety Administration]. 2015. Critical reasons for crashes investigated in the national motor vehicle crash causation survey. Traffic Safety Facts Crash Stats Report No. DOT HS 812 115. Washington. Online at https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/ 812115. NHTSA. 2020. Automated vehicles for safety. Washington. SAE International. 2016. Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles J3016_201609. Warrendale PA. State Council of PRC. 2015. Made in China 2025. -Beijing. Online at https://english.www.gov.cn/2016special/madeinchina2025/ . Strategic Headquarters. 2013. Declaration to Be the World’s Most Advanced IT Nation. Tokyo: Strategic Headquarters for the Promotion of an Advanced Information and Telecommunications Network Society. Online at https://japan.kantei.go.jp/policy/it/2013/0614_declaration. pdf. Toh CK, Sanguesa JA, Cano JC, Martinez FJ. 2020. Advances in smart roads for future smart cities. Proceedings, Royal Society A 476(2233):20190439. US DOT [US Department of Transportation]. 2010. Achieving the Vision: From VII to IntelliDrive. -Washington. Online at https://www.itsva.org/wp-content/uploads/2017/07/2010_ Pol_IntelliDrive.pdf. US DOT. 2016. Federal Automated Vehicles Policy: Accelerating the Next Revolution in Roadway Safety. -Washington. Online at https://www.transportation.gov/sites/dot.gov/files/docs/ AV%20policy%20guidance%20PDF.pdf. Yu H, Luo Y, Shu M, Huo Y, Yang Z, Shi Y, Guo Z, Li H, Hu X, Huan J, Nie Z. 2022. DAIR-V2X: A large-scale dataset for vehicle-infrastructure cooperative 3D object -detection. Proceedings, IEEE/CVF Conf on Computer Vision & -Pattern Recognition: 21361–70. [1] We use AD to refer to automated vehicles that do not communicate with pedestrians, other vehicles, roads and traffic, or the cloud. [2] The updated Chinese and English version will be published in March 2023. [3] WHO, Road traffic injuries, https://www.who.int/news-room/fact-sheets/detail/road- traffic-injuries [4] UN General Assembly resolution A/RES/74/299 on Improving Global Road Safety, 2020, https://digitallibrary.un.org/record/3879711?ln=en [5] https://thudair.baai.ac.cn/index [6] The AIROS operating system supports Apollo AIR, an AD project based on smart roads only (no dependency on smart vehicles). AIROS source codes and documentation are available in Chinese at https://gitee.com/ZhiluOS (Zhilu means “smart roads” in Chinese). About the Author:Guyue Zhou is an associate professor, Institute for AI Industry Research (AIR), Tsinghua University. Guobin Shang is vice president, Smart Transportation, Baidu. Ya-Qin Zhang is a chair professor and dean, AIR, Tsinghua University, and chair, Apollo Industry Alliance.