Autonomous Vehicles: Moving from Proof of Concept to Production

Autonomous Vehicles: Moving from Proof of Concept to Production

After a two-year hiatus from in-person events, Automotive IQ recently held its 11th annual autonomous vehicles summit in Santa Clara, California. The two-day event was fairly small, but included many of the major technology players in autonomous vehicle (AV) technology today, including IT vendors like Dell Technologies, digital providers such as trucking automation software company Locomation, and automotive industry enterprises like Ford. The focus of the event was on the hindrances in moving from proof-of-concept AVs to production vehicles, and the steps that must be taken up and down the stack to get there.

State of the autonomous vehicle industry

Krish Iyer, strategy lead for autonomous vehicles and smart manufacturing in the office of the CTO at Dell Technologies, opened the summit with a rundown on the state of the industry. He started by declaring that autonomy will probably be the most challenging engineering issue of our time.

He then ran through some hypothetical future press headlines from several years down the road. They included a headline about parallel parking not being a requirement for getting a driver’s license, human drivers having to eventually pay a fee for the luxury of driving cars themselves, and being able to automatically renew your ‘auto subscription.’

How Consumers Feel About Autonomous Vehicles

Iyer acknowledged that the future of autonomous vehicles – robotaxis, autonomous fleets – was supposed to be now. Why isn’t it?

He stated that some effects from the COVID-19 global pandemic have slowed development. Auto sales dropped and manufacturing plants closed down, leading to innovation taking a back seat. As the economy rebounded, demand for new cars, and the chips included, far outstripped supply, leading to shortages that further slowed recovery and hampered innovation.

The pandemic isn’t the only factor. For robotaxis, many believe that infrastructure connectivity and communication with AVs are requirements. That has lagged. Public perception of AVs (which could be defined as skeptical) is another major factor. According to the 451 Research’s Consumer Population Representative Survey, less than 15% of consumers say they would be comfortable with the type of autonomy that driverless robotaxis would require.

Importance of redundancy in sensor suites

Raj Rajkumar is co-director of the General Motors-CMU Autonomous Driving Collaborative Research Lab at Carnegie Mellon University, and was part of the 2007 team that won the DARPA Grand Challenge. Rajkumar spoke of how crucial it was to have sensor redundancy in developing AVs for production.

Tesla CEO Elon Musk has famously eschewed radar and LiDAR to focus primarily on cameras for perception, a move that has drawn questions about the safety of the company’s assisted driving features. Rajkumar argued that each sensor type has its strengths and weaknesses, and only by combining them can AVs get a consistently complete picture of their surroundings. In addition to the three perception sensors already listed, Rajkumar added infrared and microphones.

Rajkumar also argued that AI can only do so much. Without sensor redundancy, there will be gaps in perception that AI will not necessarily be able to cover for.

From proof of concept to production

Successful proof-of-concept tests and beta projects for AVs have been around for years. In instances where a safety driver is aboard to take over, engagement percentages (how much the AV system did the driving) are frequently north of 99%, while videos of these autonomous drives and tests are common on YouTube. Despite that, AVs have not been massively deployed. What will it take to get there? In a panel at the event, several players in the AV industry gave their takes.

Arne Stoschek, project executive for machine learning and autonomy at Airbus Acubed, raised the issue of consumer perception. He and others intimated that consumers are inherently less trusting of automated driving systems controlling vehicles than they are of human drivers. This despite the fact that tens of thousands of people die in motor vehicle crashes every year, and nearly 95% of them are due to human error. Nonetheless, the fact remains that consumer acceptance of AVs will come when it is demonstrated they are orders of magnitude safer than human drivers. How high that multiple must be is up for debate.

Steve Kenner, chief product and safety officer at autonomous trucking tech company Locomation, mentioned some of the technological difficulties of getting to production. AV test drives with 99% engagement look great, but what a driverless AV system does in those 1% of circumstances is what really matters. Getting from that 99% to 100% becomes increasingly difficult, given it encompasses the management of rarer and rarer edge cases that AV software has little to no experience managing, either in simulation or production.

Locomation’s strategy now is ‘platooning,’ or what it calls convoying. That involves two human drivers in two long-haul trucks. The lead driver manually operates the lead vehicle while the second vehicle follows autonomously so the driver in that vehicle can rest. The vehicles then occasionally switch so that the lead driver becomes the rider in the following vehicle and can rest.

Want insights on smart car technology and automotive industry trends delivered to your inbox? Join the 451 Alliance.