Dmitri Dolgov is the CTO and VP of Engineering, Waymo.
Dmitri briefly reviewed the self-driving car work ten to 15 years ago from the DARPA grand challenge to today.
They initially chose 10 driving routes and driving about 100,000 miles to teach and test its AI to get to the point of a workable prototype. This took two years.
Waymo now has a commercial self-driving car service in Arizona.
In 2017, Waymo started having totally driverless cars operating on public roads.
The challenge of self-driving cars is dealing with driving situations without rules and to get the point of a reliable product that works at all times for all situations.
Waymo switched from computer vision and perception AI to deep learning AI around 2012. Convolutional neural networks provided a big boost to system performance such as pedestrian avoidance.
At this point we do not need new breakthroughs. Self-driving cars are here.
There will be more breakthroughs in prediction, simulation and decision making. This will enable you to go that much faster.
You cannot just have simple rules there needs to situational understanding and awareness to have safe self-driving cars.
There are four areas for commercial service. There is ride-hailing, trucking, deliveries and then self -driving personal car ownership. is starting rollout currently. Waymo is working on trucking and deliveries.
Self -driving personal car ownership will come last because of the price sensitivity for the full autonomous driving systems. Note: Elon Musk and Tesla disagree and Tesla is trying to roll out full autonomous self driving systems to personally owned cars.
There was a recent talk by Dmitri three weeks ago at OFC.
SOURCES- Live reporting by Brian Wang of Nextbigfuture.com at EmTech Digital 2019 and a Youtube talk by Dmitri Dolgov is the CTO and VP of Engineering, Waymo at OFC.