Answers to Burning AV Questions
March 22, 2021—Can autonomous vehicles interpret hand signals from pedestrians? Do their self-driving features still work in the rain? What about on gravel roads?
Some of these questions were answered, and some were left unclear during a recent expert panel session hosted by Partners for Automated Vehicle Education.
Oliver Cameron, co-founder and CEO of Voyage, a self-driving car company, and Anuja Sonalker, founder and CEO of STEER, a company that has developed level four autonomous parking technology, addressed burning questions submitted by the audience and ADAPT caught it all.
Here’s what’s keeping today’s drivers awake at night.
Q: How can people have confidence in the technology, given that it is so complicated?
A: “Fully autonomous driving is very complex,” said Sonalker to begin the conversation. In order to give the public confidence in the technology, Sonalker said they have to experience it for themselves, in a positive user experience.
“It’s important to gain trust for those who will interact with the technology,” she said.
Q: Motional, a self-driving car company, asked “What parameters would you use to construct a repeatable objective that assesses safety?”
A: “I believe increasingly regulators will play a role in giving consumers that trust that this [technology] is safe and ready,” Cameron said. Currently, he believes regulators are unsure of what their role should be when it comes to autonomous technology.
“I know Elon Musk likes to make the argument that regulation is holding us up,” Cameron said, “But, anyways.”
Q: Machine-learning is often said to be inscrutable, how do AV developers deal with codes when developing these systems?
A: Cameron said this may have been true five or six years ago, but that is no longer the case.
“Now there is a robust set of tools to inspect the systems both pre- and post-deployment,” he said. “Machine-learning has proven to be far better performing with computer vision than non-machine-learning methods.”
Q: “How do you ensure your sensors will continue to operate when there is mid and dirt and bird poop?” Ed Niedermeyer, communications director for PAVE, asked.
A: Sonalker said sensors are usually protected by way of shield or covering to withstand the elements, but should an issue arise, there is a plan in place.
“When vehicles travel and one sensor is blocked, we can measure that,” Sonalker said. “We can measure if the vehicle can continue on or if it has been compromised.”
In the event that the sensors are so occluded that the vehicle’s autonomous features have to be shut off, Sonalker said the driver will be notified before the systems disengage, giving them time to take control of the vehicle.
Q: How ready are AVs to interpret hand signals from pedestrians, crossing guards, and law enforcement?
A: Cameron said with computer vision, AVs are well-equipped to be able to understand hand signals from pedestrians, crossing guards, and law enforcement.
For example, if a stop light is down and a cop is directing traffic, Cameron said an AV would have no trouble navigating based on hand signals alone.
“But for more complex situations, if it comes down to eye contact or something ‘very human,’ an AV will struggle to reason in real-time,” he said.