Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
114 REPORT APPENDIX B: ENABLING TECHNOLOGIES The following sections provide information on the technologies that can support AV/SAV operations. SYSTEMS ENGINEERING Integrating a system, or more accurately a system of systems, in the AV ecosystem requires a rigorous yet adaptable approach to the discipline of systems engineering. Systems engineering encompasses the lifecycle of the interconnected physical, cyber-physical, and social systems that make up the AV ecosystem. Considerations for quality, functionality, usability, performance, serviceability, capability, ease of installing, maintainability, and documentation are important to the safe and effective operation of the AV systems throughout the product lifecycle. One approach, Model-Based Systems Engineering, provides rigor to systems engineering and has been adopted by the automobile industry (Hart, 2015). Still, several features of the technology challenge existing frameworks including: ⢠Unstructured operating environments leading to incomplete systems requirements; ⢠High-order, non-deterministic, non-intuitive software including machine learning algorithms; ⢠Safety-critical operation and need for reliability and fault tolerance; ⢠Integration of numerous interrelated systems being developed and tested by different groups; and ⢠New model for maintainability (e.g., over-the-air updates). Adequately addressing the systems engineering challenges for AVs may be the most influential factor driving the timeline of AV deployment. System complexity is only expected to increase with expanded functionality. Integrating technical and social systems will require an effective way of communicating a level of confidence to ultimately reflect the effectiveness and efficacy of the integration and performance on these systems (Hutchins et al., 2018). SENSORS AND PERCEPTION ALGORITHMS Sensors act as the âeyes and earsâ of the AV, and perception algorithms synthesize the sensory information to develop an operational map that includes features, such as drivable area of the roadway, objects (e.g., other vehicles, pedestrians, bicyclists), and events (e.g., status of a traffic signal). Typically, a driverless vehicle is equipped with a sensor package that embraces all needed combinations of sensors such as forward-facing cameras, radars, ultrasonic sensors, lidars, and Inertial Measurement Unit (gyroscopes and accelerometers) with a priori high definition (HD) maps. The importance of accurate and reliable sensory information is driving innovations in sensor technology. These new technologies either 1) enhance the current sensor units by improving their range, accuracy, and robustness; or 2) develop completely new sensor types with smarter and smaller units. A key safety enabler for driverless vehicles is a sensor package (sensors and perception algorithms) that is providing layers of redundancy to one another and is able to combine the
115 REPORT sensing capabilities of multiple sensors, resulting in more reliable and robust perception with a broad sensing scope. The perception algorithms would frequently cross-check the data from different sensors to increase the chances that no object is left undetected and to eliminate false positives. The underlying perception algorithms for processing the data are more advanced and are capable of performing complex sensor fusion calculations and applying AI algorithms in real time. Advances in microprocessors are improving the size, energy consumption, and speed of perception systems. These advances are driving changes in-vehicle electronic architecture, such as enabling multi-domain controllers that process information from all sensors rather than using dedicated electronic control units for each sensor. HIGH-DEFINITION MAPPING HD maps are a key safety enabler for higher levels of automation (L3, L4, and L5). HD maps and their usage in localization are the most influential factors for the safe navigation of AVs. Taking the high-profile Tesla incident as a case study (Lee, 2018), the Tesla driver was killed when his AV collided with a concrete road barrier on a mountain view highway. With Autopilot engaged at the time, the system struggled to identify the correct trajectory and hit the barrier head-on. Although the road markings were faded with limited machine visibility, this accident could have been prevented if an HD map was used (Dooley, 2018). HD maps are key enablers for safe AV deployment since they assist the AV in gaining the needed contextual awareness of the surrounding environment, beyond the sensorsâ range (Figure 20), even during severe weather conditions (Jung et al., 2018). Figure 20. HD Map Assisting and HAV to Gain Contextual Awareness Beyond the Sensors Range Source: Aaron Jacob, 2019 Dynamic service is an intrinsic feature of HD mapping, where any changes occurring on the road, such as new sign added or change in road marking, are reflected in the HD map in real time. As a result, HD maps would typically function as an on-cloud server that keeps updating the map at high frequency. Vehicles would need to access this server on real-time bases and at
116 REPORT low latency. Once the AV has access to the HD maps, it applies sophisticated matching algorithms on data collected by its sensors to locate its position with high accuracyâto the nearest 10 cmâin a process known as localization (Jo, Kim, and Sunwoo, 2018). AVs typically localize themselves using roadway features and landmarks, such as marking, gantry signs, control signs, and edges. HD maps provide unique opportunities for augmenting the vehicle onboard sensorsâ limitations. HD maps coupled with perception sensors become the primary source of reference once a human is no longer in control. They also support navigation in adverse weather conditions when heavy rain or snow make road marking difficult for the machines to detect (Dooley, 2018). Figure 21 summarizes the common layers to be included in a typical HD map (Lyft, 2018). It is worth mentioning that the real-time layer is the only layer that is updated in real time while the map is in use by the AV serving a ride. It contains real-time traffic information, such as observed speeds, congestion, and newly discovered work zones. Figure 21. Typical HD Map Layers for AVs Source: Lyft, 2018 Private sector companies (e.g., HERE, TomTom) are developing services that provide a priori HD maps by collecting and developing extremely high-resolution maps to be used in AV systems. In parallel, engineers are continuing to advance near-real-time mapping methods, such as SLAM, that integrate a priori HD map data with sensors, and inertial measurement systems. Ongoing research is working toward developing techniques to improve capabilities in challenging light and weather conditions (e.g., dawn, dusk, night, rain, fog) and changing environments (e.g., seasonal changes, construction zones). Techniques for cooperative vehicle mapping (i.e., one vehicle using the perception capabilities of another vehicle to navigate more safely its own environment) offer potential strategies to address these challenges in areas with dense vehicle traffic. However, as with all mapping technologies, the availability of processing power to handle the complex algorithms required is a critical factor.
117 REPORT SIMULTANEOUS LOCALIZATION AND MAPPING Automation requires high accuracy and reliable PNT because many safety-critical decisions depend upon positioning the vehicle within the surrounding environment and accurate timing to coordinate with other systems. However, with limited rights-of-way space the critical need for rapid processing speed, researchers and product manufacturers must continue to develop PNT systems that are smaller and faster. These systems are evolving to incorporate new data sets, such as wheel speed encoder odometry, visual odometry, and other on-vehicle sensors. Additional PNT technologies are also being developed to serve as redundancies in situations when/where the Global Navigation Satellite System is unavailable or non-responsive (e.g., in areas that experience visual blocks or multiple routes). SLAM is one such solution that integrates inputs from inertial measurement units (e.g., accelerators and gyroscopes), and sensors (e.g., cameras, LIDAR, and radar) based on dynamically updated confidence in the accuracy of each of those sources. Each technology complements the other. OPERATIONAL DESIGN DOMAIN (ODD) ODD is an operating conditions under which a given driving automation system or feature thereof is specifically designed to function including, but not limited to, environmental, geographical, and time of day restrictions, and/or the requisite presence or absence of certain traffic or roadway characteristics (SAE, 2018). ODD is typically defined by the automated driving system (ADS) technology developer and original equipment manufacturer (OEM). Analyzing and understating the ODD will help provide a better understanding of how ADS features interact with and rely on various infrastructure elements. Providing an accurate model for the ODD that is matching the actual capabilities of the automotive system is a key safety enabler for any ADS feature. Also, equipping the automotive system with a reliable sensor package (sensors and perception algorithms) that is capable of detecting operations of the ADS feature outside the ODD, once they occur, is imperative for maintaining a safe operating ADS feature. For instance, the weather remains a formidable challenge for sensors and explains why many pilot programs take place in temperate climates. Rain, fog, snow, and ice can reduce visibility, obscure road markings, and reduce traction to the road. Companies are researching strategies to overcome these challenges. An ODD model would specify the weather conditions for the safe operation of the vehicle and the supporting sensors should be capable of detecting the weather conditions once they occur beyond the domain. Besides the safety benefits, such an ODD model would allow users, state, and local agencies to be able to estimate the amount of time an ADS feature is expected to operate which would assist in quantifying the safety benefits. PHYSICAL INFRASTRUCTURE The IOO would need to ensure the roadway element can be clear for its intended user to interpret. Consequently, the car manufacturers providing the ADS would need to ensure the platformâs perception suite can effectively interpret the physical element in order to safely conduct its operation.
118 REPORT The geometry, color, reflective properties, and connectivity of traffic control devices and roadways impacts the performance of ADS perception systems. The geometry of the roadway can limit sensor line-of-sight. The color of signs can impact detection, especially if it is resembling foliage or other typical backgrounds. Traditionally, cameras are used to detect the edge line for ADAS features such as lane departure warning and object detection (Howard, n.d.; Dagan et al., 2004). Perception performance of pavement markings are influenced by retro-reflectivity, contrast ratio, and width, as well as lighting conditions, road weather, shadowing, and vehicle speed. At night, pavement marking retro-reflectivity is the most important factor in detection confidence, and contrast ratio is the least important (Davies, 2016). In daytime conditions, contrast ratio is the most important and retro-reflectivity has little impact on detection confidence (Davies, 2016). The width impacts the amount of light returned to the sensor and can achieve the same light output with lower retro-reflectivity (Davies, 2016). Roadway markings are easier to see with higher retro-reflectivity at night and high contrast during the day. In order to support safe ADS integration, IOOs ability to adequately maintain its infrastructure must be considered. This includes considering the current state of roadways and signs (i.e., deterioration due to weathering or vandalism). Information about infrastructure inventory and conditions may be available to IOOs from AVs and digital maps, this information can help inform and prioritize maintenance needs. It should be noted that researchers are discovering that AI sometimes mis-interpret signs and thus perform functions incorrectly. In one example, two strategically placed stickers on a stop sign made the vehicles see it as a 45-miles per hour (mph) speed limit sign. There is a field of study called adversarial AI that aims to make artificial intelligence a little smarter. Temporary traffic conditions â such as work zones, road closures, and dynamic message signs â are not easily navigated by ADS without prior information about lane geometry, speed, and other driving information. Temporary traffic patterns, such as work zones are a major safety challenge for ADS-equipped vehicles as the roadway layout is altered making priori mapping information inaccurate thereby reliance on onboard sensing. Human drivers have the perception needed to interpret road signs allowing them to navigate these areas. However, ADS-equipped vehicles may not have trouble interpreting a new environment correctly, and as such may have difficulty navigating through these areas. Due to these difficulties, consideration needs to be given to future, design, implementation, and operations of temporary traffic patterns. Data sharing of temporary traffic information to ADS can accelerate safe deployment. Road weather remains a persistent obstacle to endure when factoring operations or ADS, so timely and accurate information about road weather conditions is important. COMMUNICATIONS AND SECURITY Connectivity through roadside units13 (RSUs) and cellular devices can improve road safety and ADS performance (i.e., improving localization accuracy or providing collision warnings). While 13 Roadside units, or RSUs are transceivers that can be mounted on a vehicle or infrastructure element or carried by hand to communicate information including signal priority requests and vehicle movements.
119 REPORT technology companies continue to research and develop ADS technologies, IOOs might consider activities that aid the deployment of such technologies. For example, IOOs may look to modify or update infrastructure design standards and/or policies, maintenance operations, and planning processes. As IOOs make investment decisions that impact these activities, an understanding of most likely ODDs and their impact on infrastructure will help them prioritize among locations, infrastructure elements, and roadway types. To enable safe operation of ADS, IOOs could help make HD maps more dynamic and accurate by playing an important role in sharing data. This data can constitute pavement markings and signage updates, work zone locations, incidents, road closures, and weather impacts. State agencies could also potentially share data (such as the location of potholes and black ice) that is collected by their equipped maintenance fleet vehicles to help update HD maps in real-time. To support connectivity the vehicle should embrace a vehicle onboard unit14 (OBU) integrating vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication within the vehicles through the OBU. This would provide opportunities for the vehicle to receive real-time dynamic data for weather, work zones, and traffic. AVs require robust, reliable, and secure communications infrastructure, but it is still unclear what communications networks and protocols will support these needs. Technology advances are transforming vehicles into powerful data collection devices that gather information from other vehicles, nearby physical infrastructure, and cloud-based data storage. Vehicles can feed this information back to drivers, passengers, OEMs, software developers, and public agencies to assist in safer navigation and better traffic management. Some AV applications will require high bandwidth streaming of data, while others require low-bandwidth, low latency, and highly secure data transfers for safety-critical applications. AVs will also rely on spectrum for high- performance radar, laser, LIDAR systems, and satellite communications for location detection and mapping. As a result, ensuring access to enough bandwidth is critical. The explosion of data transmission between vehicles, infrastructure, public agencies, and the cloud brings legitimate concerns regarding data security and network integrity. Standardized approaches to security are needed, such as encryption of all cloud communications and robust authentication schemes. These approaches are a necessary element to a secure and functional AV system. The USDOT worked with transportation stakeholders to solve the technical and governance issues for a Security Credential Management System15 (SCMS) needed to support DSRC technology, such as proof-of-concept requirements and specifications (USDOT, 2016). The USDOT partnered with the automotive industry and security experts through the Crash Avoidance Metrics Partnership to design and develop a state-of-the-art security system to enable users to have confidence in one another and the system. However, by 2018, the USDOT stopped supporting the development of the SCMS. This effort is instead being largely led by the private sector and automotive OEMs. The USDOT does however still support certain specific aspects of the SCMS, such as research regarding misbehavior detection. 14 On board units (OBUs) are devices installed in vehicles that allow vehicle and travel information to be collected and shared. 15 The Security Credential Management System (SCMS) can help improve security for V2V and V2I communications by using encryption and certificate management to protect messages.
120 REPORT Companies including Tesla and HERE are implementing over-the-air (OTA) updates, which will require secure connections to upload software and firmware updates to vehicles. The International Telecommunications Union (ITU) has developed high-level security requirements supported by a reference model and threat analysis for OTA software updates (Nakao, 2016). An AV fleet will need to be able to communicate with other vehicles, traffic signals, and other connected technologies (e.g., sensors). Additionally, vehicle communications technology may enable data collection opportunities that an AV fleet could provide to the public sector. Examples include probe data for infrastructure condition assessment or emergency services. HUMAN FACTORS A high level of trust must exist between the user and the technology to enable vehicle automation adoption. AV developments today require an emphasis on the technologies that enable the smooth transition between human and automated control. Since SAVs are anticipated to operate at L4 automation, there may be no need to consider in-vehicle driver related human factors, however L4 SAVs will likely be operating in environments with both automated and legacy vehicles. The human driver factors in these mixed environments need to be investigated. For example, communication with external road users may be harmonized between SAVs and legacy vehicles to reduce confusion. This is a topic under discussion by the SAE International Signals and Marking Devices Standards Committee (SAE, 2018). In May 2019, the committee published a standard that provides guidelines for the use, performance, installation, activation, and switching of marking lamps on ADS-equipped vehicles (Signaling and Marking Devices Standards Committee, 2019). Some companies are considering deploying SAVs with a remote-operator takeover feature that may be used when the vehicle encounters a situation that it cannot navigate, for example, a disabled vehicle in the only lane of travel. Remote operation may be implemented in many ways, and technology developers are considering issues, such as communications latency and human factors limitations.
