This is the article written by Automotive Intelligence team in KPMG Consulting.
The team offers insight into automated driving system-dedicated vehicles (ADS-DVs) based on computer vision and sensor data, suggesting a new type of social infrastructure in preparation for the forthcoming commercial deployment of automated vehicles.
The SAE Industry Technologies Consortia defines them as “automated driving system-dedicated vehicles (ADS-DVs)” , which can handle a wide variety of circumstances on the roads without drivers.
Table of contents
1. Data recording by ADS-DVs
Data recording by ADS-DVs plays a new role in reconstructing events when ADS-DVs are involved in collisions. For event analysis, the primary differences between ADS-operated vehicles and conventional human-operated vehicles lie in technology-related factors.
These factors, including sensing, processing, and control systems, will be important in event analysis. Existing methods to determine human inattention, disobeyed traffic control devices, and attempted avoidance maneuvers are sometimes insufficient to figure out factors related to ADS control systems and saliency determination. Additionally, human eyewitnesses to the events are not perfect or complete due to the limitation by the field of view, memory loss, bias, and attention.
“AVSC Best Practice for Data Collection for Automated Driving System-Dedicated Vehicles (ADS-DVs) to Support Event Analysis AVSC00004202009” (hereinafter, “the Best Practice”) provided by SAE International recommends data collection (including data elements, recording interval, and recording frequency) to support analysis related to ADS sensing, saliency determination, and vehicle motion control in the event of a collision.
The Best Practice seeks to achieve two goals:
(1) Provide information about what the ADS "witnessed" and "performed" (2) Identify the technology-relevant factors which related to the event |
The data collection by ADS-DVs could go beyond the current crash reconstruction recommendations and regulations. This is important not only for crash reconstruction but also for system performance investigations and event analysis.
Collected data are useful for:
|
ADS developers and manufacturers should prioritize data elements among limited resources and data recording considerations for event analysis to accommodate situations where not all prioritized data can be recorded (e.g., vehicle damage due to collision forces causing data loss). Vehicle design, including vehicle architecture, data collection systems, and power systems, should prioritize the collection and availability of these elements after the collision for event analysis purposes.
2. Data prioritization and categorization on data collection
Data elements should be prioritized based on their anticipated value to event analysis. Not all data can be recorded, as there could be significant variation in the interval over which data elements are recorded due to how data is buffered and written. Methods for buffering and writing data should be considered for trade-offs between size, writing time, and value to event analysis for each record.
For example, it may take the same amount of time to record eight records of some data element and three records of another due to size differences. ADS developers and manufacturers would get a more complete record of a certain data element before recording other data elements within the same priority level. The approach to prioritizing data elements will vary depending on the operational design domain and use case.
Data elements are identified and defined including units, the minimum of resolution, range, accuracy, recording frequency, and recording interval, and prioritization order. Data collection practices will evolve with ADS technology, and considerations are required related to topics such as off-board storage, availability of vehicle subsystems (e.g., sensing, communications, and power systems), changing vehicle architectures, and compliance.
The Best Practice recommends 39 data elements into four categories based on their contribution to event analysis.
Table 1 shows the selected examples of the 39 data elements:
[Table 1: Selected examples of the 39 data elements into four categories]
Vehicle control (“What the ADS did”) |
|
---|---|
Saliency (“What the ADS thought was important”) |
|
Sensing (“What the sensors saw”) |
|
General Parameters |
|
Source: Prepared by KPMG Consulting Co., Ltd. based on SAE International “AVSC Best Practice for Data Collection for Automated Driving System-Dedicated Vehicles (ADS-DVs) to Support Event Analysis AVSC00004202009”
Each ADS is unique, and each collision is different; therefore, certain data elements should be prioritized along with each system or each event. All data elements are recorded on an as-applicable and as-available basis. If an ADS determining a sensor is blocked by something flying and the ADS still operates in a degraded mode, recording the sensor data is prioritized as it helps explain system performance. Or if a power grid within the vehicle is in failure, it results in fault code logging with no recording capability.
3. Design data collection and storage
Designing operations for data buffering and writing involves trade-offs between recording different data elements. These trade-offs depend on factors such as space efficiency, the location of systems (on-board or off-board), data origin, and relevance to specific events. Event-recorded data can be stored in on-board systems, transferred to off-board systems (e.g., cloud storage), or both. The interval and frequency of data recording are also influenced by storage capabilities and data prioritization.
Data collection performance can vary between a fully functioning vehicle and the one operating in a degraded mode, such as under minimal risk conditions. In the degraded mode, the vehicle may lose certain capabilities, affecting data capture, processing, and storage. Vehicle malfunctions or damage from collisions could limit data availability. The completion of records depends on the availability of sensing, communication, and power systems. Disruptions to power or communication systems may result in the lack of data recording.
If ADS developers and manufacturers understand data collection needs better, vehicle technologies and architectures will evolve accordingly. Existing vehicle architectures may impede the access to power and communication systems and pose challenges based on the location of these systems within the vehicle.
4. Evolutions along with development on ADS as public transportation
Data from ADS developers and manufacturers offers objective, reliable factors that build public trust. It is also important to have diverse sources, including event data recorder (EDR) data, human eyewitnesses, and scene analysis, to determine key factors associated with an event.
As ADS-DV models evolve, the data used to assess safety improvements will change. However, collision reconstruction for event analysis requires certain data based on requests from authorities, who may not be familiar with data from ADS-DVs. Some data should be acquired promptly in cases of hit-and-runs or injuries.
Authorities should be informed about the unique nature of ADS-DV data for secure public transportation. ADS-DV operators should design their systems related to data flow and storage to provide necessary data in an adequate format or visuals at the best timing to improve road safety as a part of the social infrastructure.
Author
KPMG Consulting
Manager, Yukie Koyano
Connect with us
- Find office locations kpmg.findOfficeLocations
- kpmg.emailUs
- Social media @ KPMG kpmg.socialMedia