Self-driving vehicle designers continue to wrestle with concerns over safety and ethics, even as driving systems like automated emergency braking and other applications have been deployed in some newer vehicles for years.
They worry over a long list of concerns, including how autonomous vehicles should react if, for example, a pedestrian should suddenly step into a roadway where cars are AVs zipping by.
A panel of designers and engineers at Sensors Converge 2024 this week raised the question of how an AV would or should react if a 6-year-old girl wandered into roadway from the left and and an 80-year old an walked into traffic from the right, leaving too little room to brake in time. It turns out that insurance companies have been mulling for years over where liability would apply in such scenarios and whose life and health would be sacrified: the old man or the girl? An impossible choice that some on the panel said would favor the girl over the old man, given he has already seen his life.
Using AI to tell the vehicle to swerve might not meet the standard most humans would accept, however. “Ethics with AI is always a hot topic and it is a hard problem. The logic of social economics is not the logic people accept,” said Sabrina Yuan, vice president of strategy and business development at Cariad, a division of Volkswagen. “There are many issues we haven’t answered yet.”
Another example the panel discussed is how some AV design teams have discussed keeping speeds of the vehicles slower in neighborhoods with expensive homes, in an attempt to find the least amount of monetary damage in the event of an accident. Ethical challenges will vary fromplace to place, and such concerns makes it challenging for engineers.
Such challenges have already slowed down the broad production of self-driving vehicles at SAE levels 3 and 4 where no driver need be present for a vehicle to operate. Some experts believe widespread production rollouts won’t happen any sooner than in a decade or more and public adoption will be even further off.
AVs will be driving alongside standard vehicles for decades to come, partly because the average age of cars on US roadways is more than a dozen years, Yuan said, meaning human drivers will interact with AVs for decades to come.
Some accidents with Teslas have traced to the way drivers interpret the capabilities of their vehicles, panel members noted. Tesla instructs drivers to stay in control of their vehicles in full self-driving mode, but drivers believe their cars will perform better than actually do, panelists said.
One solution will be for auto makers to do a better job educating drivers on the limitations of their AV systems. Some auto makers are even using the lessons of Generative AI and are attempting to turn their vehicles into digital assistants.
GenAI and personal car assistants will conceivably help car owners be able to talk to their cars about a car’s maintenance needs and performance, said Eric Reuthe, vice president of engineering for Zapata AI. “With personalization on board the car, I’m excited about the future of diagnostics, instead of a mechanic using a port reader, you’ll ask the car what’s the problem,” he said.
“Hardware is progressing and now we have more computer power that is closer to the edge. With tinyML and anomaly detection, IoT can be on board,” Reuthe said.
Zapata.AI works with the Andretti Formula One racing team to perform predictive failure analytics on race cars as they travel at speeds well over 200mph. A truck that moves from race track to track with a server inside ruinning dual Nvidia A100 GPUs to interpret data from dozens of sensors on each race car. The data is transmitted often over 5G, but a connection to the cloud is not necessary except for some training and retraining of models in the cloud.
Even with the advances of Zapata.AI and Andretti, Erik said a future where passenger cars on public roadways can interpret data away from the cloud is uncertain. “Would infrastructure be ready for this?” Reuthe mused.