2021 Gaps and Recommendations

In the last years, a big effort has been dedicated by many companies, associations and universities in standardisation of Connected and Automated Driving (CAD) vehicles. The number of published standards have increased year by year reaching its peak in 2018 (figure 1). This indicates an increasing attention on standardisation activities by different stakeholders since 2016.

At the end of 2020, less than 200 standards in the field of connected and automated driving have been published, and there are still around 40 standards underdevelopment. An impressive number of standards have been published so far (figure 1). However, there are still some gaps that we want to highlight in this article. The gaps represent the basis for making final recommendations.

To help the readers all recommendation are marked in bold throughout the text.

On the web page Standard Collection, all the CAD standards are listed. They have been distinguished in two categories: underdevelopment and published. Furthermore, the standard collection has been divided in domains.

The domains have been chosen before starting the research and they are listed below:

• Terms & Definitions
• Management/ Engineering Standards
• AD/ADAS functions
• Testing, Verification & Validation
• In-Vehicle Systems, Networks, Data and Interface Definition
• Connectivity
• Human Interaction
• Artificial Intelligence
• Safety
• Privacy & Security
• Map and positioning

In the figure 2, we have graphed the number standards per each domain, divided into published and underdevelopment.  From the graph, three indications appear clearly:

1. Connectivity is the area with the highest number of standards. This result is completely expected because no communication can be established without a standardisation of interfaces, bandwidths, frequencies, protocols, etc.
2. Testing, Verification and Validation is the only domain where the number of standards underdevelopment is higher than the ones already published. The reason behind could be related to previous gap analysis [1] [2], that highlights a gap regarding this key doman CAD deployment.
3. Artificial Intelligence (AI) is the only domain in which we expected some standardisation  activities, but at first glance a lack of standards related to the field of connected and automated driving has been recorded.

In particualar, the lack of AI standards suggested that further investigation are needed. In the following section, the topic of AI will be analysed.

Standards for Artificial Intelligence

From a further investigation of AI standards resulted that the International Standard Organization (ISO) decided in 2017 that the Joint Technical Committee “Information Technology” should create a subcommittee (SC42) on the topic of the Artificial Intelligence. The subcommittee split the job between several working groups:

• WG1: Foundation standards
• WG2: Big Data
• WG3: Trustworthiness
• WG4: Use cases and applications

All the standards in preparation by these working groups are not specifically developed for connected and autonomous driving, but they can apply to a wide range of applications.

In the automotive industry the standards related to functional safety (ISO 26262) and SOTIF (Safety of the Intended Functionality, ISO 21448) have started to consider the problem of introduction of artificial intelligence algorithms in the ADAS and autonomous driving systems.  However, to the best of our knowledge the safety problem related to the introduction of AI software in the automotive applications is still open.

On the other side, the direction took by the WG3 could potentially help to solve the safety, which is strictly related to the topic of trustworthiness.  In particular, the WG3 has the following tasks [4]:

1. Establish trust in AI trough transparency, verifiability , explicability and controllability.
2. Investigate threats and risks of AI systems
3. Investigate approaches to achieve AI systems robustness, resiliency, accuracy, safety, security privacy.

All three tasks are essentials for deploying AI software in connected and autonomous driving system.

Firstly, the concept of transparency and explicability is extremely important for using artificial intelligence techniques in those software modules responsible for the final manoeuvre. In case of accident due an apparent wrong manoeuvre, the decision process took by the autonomous driving car before the accident has to be completely transparent and explicable.

Secondly, the topic of threats and risks is essential to assess ethics and goodness of choices. In particular, considering the nature of some AI techniques, the goodness of choices needs to be assessed by using a statistical approach. Indeed, the risk cannot be eliminate, but it could be reduced as much as possible. The residual risk remained needs to be quantified to understand if could be acceptable.

Finally,  the topic of robustness is very important for each software component using AI models in CAD vehicles. The word “robustness” is used as a general terms for describing a series of properties of AI techniques [4]. However, we can summarize the robustness as the property to maintain unchanged the functional performance with respect to small input variation. In detail, it is not required that outputs remain exactly the same with respect to small input variations, but the performance have to remain unchanged. As an example, considering to have an artificial neural network able to recognize cars with a detection rate of 90%. The property of robustness could require that the detection rate remains unchanged regardless of time of day.

In order to verify the robustness or the functional performance of AI models, formal methods cannot be use in practice, therefore statistical methods have to be implemented. On this regards, as suggested in [4], there is a need for international standardisation of field test for AI. Probably, this could be varying from industry and applications.  The final field test of AI application, can lead to increase the trustworthiness, the safety, the credibility of this techniques. Moreover, this standard will help to define responsibilities of different stakeholder during design, implementation, verification and certification phase.

As mentioned before, all these aspects are crucial for AI deployment in CAD vehicles. Thus, the WG3 is moving to the right direction to give a common solution to these challenges and enable automotive industries to use safe AI models.   Liaison activities needs to be engaged between SC42 on Artificial Intelligence and :

• ISO/TC (technical committee) 204 on Intelligent Transportation System
• ISO/TC 22 on Motor Vehicle
• IEEE P2846 on Autonomous Vehicle Decision Making

Standard recommendations in last years

The scope of this article is to give to readers a series of recommendations on CAD standardisation . However, it is important to mention that ARCADE is not the only consortium trying to analyse the standards published so far and to highlight the main gaps. Three organisations worked on in the same topic in the last years. They published the following documents:

• ISO/TR 20545:2017 “Intelligent transport systems — Vehicle/roadway warning and control systems — Report on standardisation  for vehicle automated driving systems (RoVAS)/Beyond driver assistance systems”, July 2017
• “Standardisation  Roadmap for Automated Driving”, Verband der Automobilindustrie (VDA), May 2019
• “Connected and Automated Vehicle Technologies – Insights for Codes and Standards in Canada”, Canadian Standardization Association Group (CSA), June 2020

In particular, the last two publications have been published recently, thus they can offer valuable information to understand the remain gaps.

ISO/TR 20545:2017

This technical report published in 2017 give to the readers some interesting recommendations, some of them have been already considered in new standards. However, some others are not considered so far. In this article we emphasize on the remain standards to be worked.

The ISO/TR 20545:2017 advises to standardise functional requirements, functional architecture and interfaces between the different components. This standardisation will help the industry to be faster and it will increase the compatibility between different solutions in the market. Until now, there is only the standard ISO/DIS 23150 that goes in this direction trying to standardize the interface between perception sensors and data fusion software modules.

Furthermore, the ISO technical report recommends the standardisation of requirements of some particular safety critical modules:

• Requirements between the driver and system at each level of automation and during the control transition
• Requirements of system operations in case of a system malfunction.
• Requirements for reliability and certifications.

Still today, several regional institutions (NHTSA, SAE, BASt) have their own definition of level of automation. Therefore, it is needed an international definition that avoid any confusion in the automotive industry and with consumers.

Finally, the ISO technical report advises to standardize the visible interface with other vehicles, cyclists or pedestrians when the autonomous mode is activated.

VDA recommendations

The technical report on standardisation road map by VDA has been published in May 2019. Since then, a lot of standards have been published or have been started the discussion process. Reviewing the contents of the document [2], we can underline some important domains that still requiring a standardisation . The recommendations have been summarized and categorized into domains:

CSA Group recommendations

The CSA published an interesting report in June 2020 [3], in which they have collects feedback from 14 stakeholders, who belong to automotive industry (OEM, technology companies), public transportation agencies, academic and research institutions and standard developing organization (SDO). Based on their feedback they have identified gaps and needs in codes and standards. They have divided that in 8 themes. We summarize, as in the previous section, the main recommendations in the table below:

Theme	Recommendation by CSA
Harmonization and interoperability	Harmonization of standards dedicated to V2X (vehicle-to-everything) communication globally. In particular, on band allocation. Standardisation  of communication in emergency situations. Standardisation  of information related to road condition and weather.
Uncertainties with enabling communication technologies	Make a clear decision on type of technology to use for the communication: DSRC or C-V2X. Lack of standards in long range wide-area applications
Compliance verification	Define responsibilities for certification of CAD technologies. It could be OEM and Tier1 responsible for the certification  (or better self-certification) or a third party entity.
Physical infrastructure	Do CAD vehicles have to adapt to infrastructure or vice versa? Indeed, it is needed a standardisation  effort on this directions. In particular for static signs, electronic signs, road works, road continuity.  The reasons behind this standardisation  activities are two: 1. Accommodate the needs of CAD vehicles 2. Harmonize the signals as much as possible between regions to facilitate the deployment of CAD technologies. It is not sustainable to adapt to CAD technologies to the specific needs of each jurisdiction.
Operation Design Domain	Lack of standards that define metrics of ODDs. It is fundamental to define the measurable conditions (temperature, precipitation, etc.) , the type of road, the marking signs, the connectivity status that allow or not allow a CAD vehicle to operate.
HD mapping and localization	Standards on map creation will help increase compatibility between different providers.
Cybersecurity and Protection of privacy	There are already a lot of activities in cybersecurity space. However, there are some concerns related to SCMS (Security Credential Management System), that manage credentials and ensures secure V2X. One is related to misbehavior detection management and the other on privacy.
Technology Maturity	The main gaps identified in this area are related to difficulty to identify other vehicles in non-optimal conditions, as well as objects with a low reflectivity or unique or less common objects (like motorcycles).  Moreover, construction zones could be difficult area to detects.   Connectivity could help to solve this problem when will be available. However, it is more likely that in the next year we assist to a natural enhancement of  the capability by using a perception system by using machine learning.

Final recommendation

The landscape of standardisation of CAD vehicles is very wide and around 200 standards have been published in different domains so far.

This article attempt to give some indications based also on the experience of other consortium. Indeed, more liaisons activities are needed to bring together a wider audience of experts on each single domain.

Nevertheless, from the study and gap analysis we can summarize our recommendations as follows:

• Standardisation of field test for AI
• Liaison activities between AI and automotive and mobility industries. In particular, we recommend a liaison between SC42 on Artificial Intelligence and :
  • ISO/TC 204 on Intelligent Transportation System
  • ISO/TC 22 on Motor Vehicle
  • IEEE P2846 on Autonomous Vehicle Decision Making
• Standardisation  of :
  • Functional requirements, functional architecture and interfaces between the different components
  • Requirements between the driver and system at each level of automation and during the control transition
  • Requirements of system operations in case of a system malfunction.
  • Requirements for reliability and certifications.
• Standardisation  on terminology in particular on level of automation
• Standardisation  of external visible interface of vehicles communicating with pedestrians and cyclists
• Standardisation  of operation for managing a fleet of CAD vehicles
• Standardisation  for recognition of emergency vehicles
• Standardisation of ethical requirements related to actions to undertake in case of an emergency situations.
• Standardisation  of a social recognized model to compare the performance of “safe automated driving functions” with respect a human driving performance. This is very useful for CAD acceptance and introduction.
• Standardisation  of requirements on minimum risk condition
• Harmonization of standards dedicated to V2X (vehicle-to-everything) communication globally
• Standardisation  of information related to road condition and weather.
• Standards in long range wide-area communication
• Standardisation of physical infrastructure (static signs, electronic signs, road works, road continuity)
• Standards that define metrics of ODDs
• Standards on maps

Bibliography

[1] ISO/TR 20545:2017 “Intelligent transport systems — Vehicle/roadway warning and control systems — Report on standardisation  for vehicle automated driving systems (RoVAS)/Beyond driver assistance systems”, July 2017

[2] “Standardisation  Roadmap for Automated Driving”, Verband der Automobilindustrie (VDA), May 2019

[3] “Connected and Automated Vehicle Technologies – Insights for Codes and Standards in Canada”, CSA Group, June 2020

[4] Thomas Zielke, “Is Artificial Intelligence ready for standardisation?”, Proceedings of 27th European Conference, EuroSPI 2020, Düsseldorf, Germany, September 9–11, 2020