The Warwick University's View on Challenges in Autonomous Drive

Siddartja Khastgir, researcher at the Warwick University, Coventry, talked with Automotive IQ on how development and testing procedures need to change with the advent of advanced driver assistance systems. Can you please tell us a bit about your current projects and the ISO Intelligent Transport Committee you are currently involved?

Currently we are looking into creating standards for various ADAS systems and explore the topics that need standardization for high levels of autonomy. We have already published a technical report (which is out for public comment) on possible areas for standardization for higher levels of autonomy.

How would software and hardware development processes need to change in order to effectively work together to ensure autonomous vehicle safety? What changes in current hazard analysis and risk management will need to be developed in order to reach those goals?

As we move from rule based behavior towards artificial intelligence and machine learning algorithms, there needs to be a seismic shift in the way we validate our software.

Validation of AI-based algorithms offers a challenge not encountered by the automotive world before. Learnings from computer science need to be incorporated in the automotive world. The shift is evident as various manufacturers are partnering with tech firms.

Current hazard analysis and risk management processes are based around driver being the last line of defense. However, with fully autonomous systems this key aspect will change. A new or a modified framework needs to be accepted and adopted by the industry. International standards can play a key role by providing guidance to the automotive community.

[inlinead]

In real-traffic testing scenarios: where do you think there’s still work to be done to help make autonomous vehicles even safer, keeping in mind all the current projects run by various organizations / institutes?

The biggest challenge with real-traffic testing is repeatability and time. For example, sunset and sunrise situations offer a challenge for vision based systems. But we have only one sunset and one sunrise a day. Therefore, it would require large amount of time to test these systems.

Another challenge is around identifying the “interesting” scenarios that should be tested. In other words, having a full fault coverage, i.e., knowing all possible faults that could occur and testing systems for them.

In your opinion, what are the biggest challenges that the automotive industry will face when ready to bring the autonomous vehicles in the mass consumer market?

On the technical aspect, the biggest challenge for the automotive industry will be around ensuring “enough” verification and validation of autonomous vehicles. This includes having the knowledge or the ability to find all possible faults which can occur in a system. This requires a robust hazard analysis and FMEA process.

Additionally, a test setup which can efficiently perform the tests in a cost and time efficient manner is needed. On the legislative side, lack of standards and legislations to guide the manufacturers is a major challenge. However, various governments are working hard to solve this issue.

Various national and international bodies have started preparing regulations for self-driving cars. Do you think there’s a danger of over regulating? Or that the political scene is not equipped with the technological knowledge to regulate such highly complex technology?

While I think there is a need for national and international bodies to have regulations in place for self-driving cars, they need to engage with the technical community for a more collaborative approach in preparing regulations.