Safety Standards, Process, and Key Learnings for Autonomous Vehicle Software

essidsolutions

The potential of autonomous driving continues to be explored. However, bringing that innovation to mass production requires a dedicated understanding and commitment to the standards within the automotive industry. Jungbae Yoon, process & safety team leader at STRADVISION, offers insight into challenges and in-depth processes for software companies entering the arena.

In recent years, the autonomous driving and ADAS tech space have found exponential interest both on a global level and in the number of major automotive and software companies becoming involved. In turn, the software behind that innovation has also evolved with every success and challenge identified. This level of exploration has resulted in a more defined relationship between software companies and the automotive industry, especially with regard to international standards for functional safety and the process toward more mass production of tech.

Some major standards and certifications important to the success of autonomous and ADAS software implementation include:

    • ISO 9001: For creating, implementing, and maintaining a Quality Management System for products and services
    • ISO 26262: For preventing accidents by the error derived from the vehicle’s E/E System (Electrical/Electronic System)
    • ASPICE:  An auto industry standard defining the framework for in-vehicle software development processes
    • SOTIF: “Safety of the Intended Function” for the absence of unreasonable risk due to hazards resulting from functional insufficiencies of the intended functionality or by reasonably foreseeable misuse by persons

The automotive industry upholds strict and specific standards when it comes to driver and passenger safety, and the above are just a few of the standards and goals put in place for the emerging software, ADAS, and autonomous driving field within automotive. However, to be a leader and innovator in that field’s safety brings the need not only to rely on automakers, Tier 1s, and OEMs for guidance and understanding but also for the software companies themselves to build an internal team and put a system in place that embraces a forward-thinking automotive understanding and unparalleled mindset for safety as the priority.

Understanding Challenges and Staying Proactive in the Process

For AI and deep-learning software companies entering an emerging space like autonomous driving, it is important to avoid being reactionary to the standards in front of you and, rather, proactively build a team that is highly informed and experienced with the automotive industry and its processes. In other words, your company must understand equally from both the software solution and automaker perspective each step of the way.

A proactive approach is also imperative when it comes to overall ongoing goals. While some ADAS and autonomous driving software have requirements to achieve specific certifications on their road to more mass production (such as ISO 26262), there are always higher standards that must continually be pursued (such as SOTIF) for the betterment of the product performance and safety overall. For example, SOTIF addresses an unexpected range of risks that ISO 26262 cannot handle, such as the current performance limitation of deep neural networks (DNNs). Even without a malfunction from the internal software, objects may be detected incorrectly due to performance limitations. SOTIF aims to cover risks brought on by these performance limitations, making it important always to identify and acknowledge the limitations of each standard, even when they are achieved.

This approach will help your company’s software evolve beyond potential industry limitations and, most importantly, help ensure optimal safety and reliability with the product. Furthermore, positioning your company as an always-on leader among standards and safety implies your software’s maturity and higher potential for mass production.

See More: AI Technology is Helping Fleet Managers Improve Safety and Increase Efficiency

Challenges and Rewards of Working with Automakers, Tier 1s, and OEMs

Those within the software always pursue relationships and work directly with automakers, Tier-1s, and OEMs. As these customers and collaborators have certain automotive industry processes and standards to achieve for their products, so too must the software being implemented in those products.

The automotive industry largely takes a more conservative approach when it comes to standardization and certifications, as the priority is always and must be safety. Some software entities new to the space may find this as a difficult adjustment in their innovations. However, working with automotive industry customers and collaborators can be both challenging and very rewarding in that it pushes software companies to their limits and beyond when it comes to innovation, safety, and standardization for a mass market like automotive.

Similarly, there have been huge developments from the automotive side where the industry is jumping into software development and collaboration for more ‘software-defined’ automotive development. Automakers, Tier-1s, and OEMs’ software knowledge are at the highest levels today. This brings great potential for collaborations that bridge innovation and safety like never before.

Introducing Multiple Methods to Educate the Overall Process

It is also important to understand the limitations or perceptions of different software development methodologies as well. DNNs have become a major learning method in the space, but there are challenges in it being a ‘black box’ outside of human comprehension – which, even with proven test results, can still lead to doubt among consumers and industry in its safety and reliability. Avoiding a one-method approach in the development and embracing multiple will aid software companies in more 360 learning, overall understanding in testing, and better industry and public knowledge and perception of their processes for safety.

Notable examples outside of DNN include AI safety and algorithms and Agile methodology as a more recent development. The former can aim for an organizational level of research and enhancement of capacity to supplement the limitations of DNNs. The latter shifts toward the need for flexibility and pragmatism in delivering the final parts/elements of the software (versus the entire application). Each of these, of course, brings its own challenges while working within the compliance and standards of a specific industry like automotive.

At the end of the day, the process of software development focuses on enhancing product quality and minimizing product error/malfunction. Having certifications in place, such as ISO 26262 and achieving those goals are hugely responsible for not only the safety but the overall performance of software looking to revolutionize an industry. Each level of compliance brings its own individual steps toward preventing things like systematic failure, predicting the edge case (hazard scenario), developing algorithms to prevent/detect multiple situations, and minimizing or at least managing the expectations of false positives during testing.

Those in-depth steps needed take a ton of time for companies to test and learn, then repeat and repeat again. By bringing in multiple methods to your software development arsenal, a company will help ensure they acquire optimal levels of data to identify all possible hazard scenarios as they pursue higher KPI levels and qualifications within the industry.

What innovative safety standards are you abiding by to aid growth? Share with us on FacebookOpens a new window , TwitterOpens a new window , and LinkedInOpens a new window .

Image Source: Shutterstock

MORE ON AUTONOMOUS VEHICLE