Chapter 8 - Leading Trends in Automotive AI

For all types of vehicles, the majority of respondents (71%) are implementing AI in their vehicles, whether AI is driving the vehicle design (24%) or affecting some components (47%). 

Region

While a majority of respondents indicated that AI is affecting some vehicle components, those in the Middle East and Africa are extensively affected by AI in their vehicle designs, while those in the Oceania region are not implementing AI in their vehicles. 

Automotive Development Focus

Breaking the responses down by automotive development focus, respondents are more extensively using AI in Dealer Management, while AI is otherwise affecting at least some vehicle components for most areas.

It matters where AI is implemented in the development process, largely because AI can continue to learn and evolve in the product, in which case it can be nondeterministic and then requires different functional safety standards like ISO/PAS 8800.

Survey respondents said that they are using AI during development, mostly during the design and coding phases, but also many in the testing phase. AI-enabled functions — especially ADAS, autonomous driving and infotainment — are predicted to boom by 2035, according to a McKinsey report, but AI also shows the potential to improve the functionality and features of other areas like powertrain and connected services. Development teams may feel competitive pressure to adopt and apply AI technology as soon as possible to meet anticipated demand for SAE levels 2 and 3.

While more automotive developers said that they use AI in development only, 45% said they are using it in both product and development, with 70% responding that the AI model is live once it is in the product. 

The use of AI during development is exciting but raises caution when it comes to compliance. (For example, MISRA checking with AI is not allowed.) This emphasizes the need for ISO/PAS 8800 compliance, especially as safety is a leading concern again this year (54%) out of all AI vehicle development concerns. 

However, while ISO/PAS 8800 accounts for the nondeterminism of AI, the product still must be verifiable, meaning that any machine learning must still meet the requirements of the product — and with 70% still using an AI model in the product, caution is advised. 
 

Among the regions more largely represented in the survey, North American respondents (84%) have the highest use of AI live in the product, compared to Europe/UK (59%) and Asia (67%). In addition, those with the most experience (64%) are applying that experience in development, using more caution about leaving the AI active compared to those with the least experience (78%). 

As AI use increases in the design and development of automotive software, various concerns arise, especially as regulations and guidance around AI are still being developed. Automotive teams are under pressure to compete in the shifting market and therefore must deliver a quality product on time — while keeping costs down and ensuring safety and security. 

Safety and “Safe decision-making for AI algorithms in autonomous/semi-autonomous vehicles” was the leading concern in AI vehicle development (54%). Of those who were most concerned with AI safety, 30% were concerned when the AI was deactivated in the product and 70% were concerned when it was left active in the product, aligning exactly with the overall results. 

Development teams who are guided by functional safety standards need to employ additional considerations when using AI, as the algorithms tend to be nondeterministic. Fortunately, existing standards are already being adapted and newer standards like ISO/PAS 8800, “Road vehicles — safety and artificial intelligence” have been introduced. In addition, techniques are already available that can be applied to AI algorithms, but there is still much work to be done to ensure and enforce autonomous vehicle safety with AI technology. Functional safety standards around AI are not developing as quickly as the software is developing, and AI introduces many variables, especially when the model is allowed to evolve. 

Security, specifically “avoiding vulnerabilities and cyberattacks with the introduction of advanced AI,” was the second most-concerning issue for survey respondents (41%). Connected systems with increasingly complex technologies like those used for AI create many more attack vectors, which malicious actors can then exploit, and this can permeate across the whole line. The types of organizations most concerned with safety were Tier 1 suppliers (32%). As Tier 1 suppliers can supply many of the same components as the manufacturer, they need to be especially careful about security risks. Similar to what is happening around safety standards, there are numerous security standards and regulations that can and are being adapted for the inclusion of AI. 

Time-to-market and development costs fell somewhere in the middle as concerning but not overly so. Once concerns about safety and security are addressed, time-to-market and development costs can be naturally reduced, especially when developers are encouraged to use static analysis tools that streamline compliance and quickly find vulnerabilities across large code bases. 

Like last year, getting accurate results for “predictive maintenance and vehicle diagnostics” and generative AI, “keeping code quality high when written by AI tools” were mildly concerning this year. It is interesting that generative AI is the least concerning for respondents, when quality was the overall leading automotive development concern in this report. It is possible that respondents are assuming that AI contributes to better quality code. According to a Stack Overflow survey, around 85% of professional developers are using or planning to use AI coding tools this year, but in automotive development especially, it is important to always have a human developer checking AI-generated code. Using a Perforce Static Analysis tool with optional AI-assisted code remediation keeps development velocity high by making accurate, contextual fix suggestions, while requiring developer approval before the changes are applied.

A Closer Look At AI/ML Use in Automotive Software Development

Respondents who are using AI/ML in automotive software development were asked about specific areas of automotive development focus. 

Most of those surveyed were focusing on AI in ADAS and IVI systems. 

Additionally, the “Other” responses (21%) give an interesting overview of the specific functional areas where AI is used, rather than just the components. In the Powertrain answer, for example, AI can be used during development and also live in the product.

Notable “Other” responses included: 

The recently published ISO/DPAS 8800 standard addresses AI-specific challenges for the functional safety for all road vehicles.

Compared to last year, there was less certainty about whether ISO/DPAS 8800 will be important to automotive software teams, even though the safety of AI was still a leading concern and most AI models are still active once it is in the product. Regionally, more automotive professionals from Africa (87%), the Middle East (67%), and North America (66%) seemed aware of its importance, while 41% in Europe/UK still don’t know whether it will be important to their teams.