How to help AVs adapt to driving norms around the world
Illustration: Rebecca Zisser/Axios
AV developers are using their test vehicles to collect data on the interactions between vehicles, pedestrians and cyclists, but that data is currently specific to West Coast cities like San Francisco.
The big picture: The cultural norms of driving vary widely from one region to the next. To operate safely and be deployed widely, AVs will need to draw on global data sets that are locally customized and continuously updated to account for both changing behaviors and new modes of transportation, like electric scooters.
Jaywalking is one example of these regional differences:
- In Japan, only 7% of people cross the street against a red light, as opposed to 67% in France.
- In big U.S. cities like Boston and New York, jaywalking is so commonplace that police officers rarely enforce laws against it.
- In Germany, most people obey the Ampelmännchen (“little traffic light man”), whereas pedestrians in the U.K. tend to be more defiant.
What's needed: Massive amounts of vehicle data have to be processed in real-time for vehicle path planning — without cloud connectivity — to ensure AVs perform safely and smoothly. At the same time, AVs also act as probes, capturing anomalous road behaviors that are then identified and processed through cloud-enabled feedback loops.
- Once an anomalous situation is detected, it is flagged and transferred wirelessly to the cloud for offline processing, including tuning of the machine learning models used on the vehicles.
- These tuned models are transferred back to AVs via ongoing system updates, ensuring they continue to improve over time.
The bottom line: Though it might be a while until self-driving cars have spread across the world, technology developers that master scalable solutions for these regional differences will pull ahead of the competition and make for a smoother transition of AVs onto human-dominated roads.
Sid Misra is the CEO and co-founder of Perceptive Automata.