Sebastian Thrun, the founder of Google X and the Google Self Driving Car Project, has shared some thoughts on Autopilot in a recent interview with Lex Fridman:
Sebastian Loves His Tesla
Sebastian starts off by saying that he owns a Tesla, and loves it. He says he uses Autopilot “literally every day” and credits it with keeping him safe on the road.
“It’s a beautiful technology”
“It turns me into a much safer driver, and I’m 100% sure that’s the case”
This directly contradicts Google’s official company story that the Waymo team originally tried a drivers assistance approach, but abandoned it because people wouldn’t pay attention. If the early members of Google’s Self-Driving car team ever did believe advanced driver assistance wasn’t safe, Autopilot has now convinced them they were wrong.
This is a critical point as the merits of Waymo and Tesla’s approaches are weighed, by legislators, regulators, and customers. No Autopilot user –– not even the founder of Waymo –– thinks Autopilot makes them less safe. Everyone who has used the technology is convinced it makes them safer. The only people who don’t like it are Tesla short-sellers who are happy to see people die as long as they can make a few bucks on it.
Deep Learning Changed Everything
The Google Self Driving Car project started over 10 years ago. For some bizare reason, the public accepts the fact that it has been 10 years since the project started and a large scale launch is nowhere in sight. Waymo is currently burning more than a billion dollars a year. But I digress.
When the project was started ten years ago, deep learning hadn’t really caught on yet. For that reason, Waymo was built on what Thurn calls “a geometric approach”, which means using expensive LIDAR sensors to accurately measure the distance to objects around you.
Autopilot, on the other hand, is based on a computer vision and deep learning based approach. The entire platform is built around fast iteration and globally crowdsourced machine learning. Thrun is essentially hinting that advances in Deep Learning made Waymo’s approach obsolete and that Telsa’s approach may be the way forward.
Is LIDAR a Crutch?
One of the biggest debates in the autonomy space is whether LIDAR is a crutch. LIDAR systems can cost $75,000+ and make self-driving cars far too expensive for an average person to own. Tesla CEO Elon Musk famously claimed that LIDAR is completely unnecessary and that Autopilot could achieve Level 5 autonomy with just cameras, radar, and ultrasonic sensors.
So what did the founder of Waymo have to say about LIDAR when asked pointedly about it by Lex Fridman?
“We all know that people can drive cars without LIDARs in their heads. […] So eyes must be sufficient. In fact, we could even put a camera out […] and we would be able to drive a car that way. So a camera is also sufficient”
Thrun makes a good point: If we can drive a car with a camera feed, why can’t a computer? This existence proof is obvious to anyone who has thought much about Autopilot, but it’s remarkable hearing it come from the founder of Google’s self-driving car project himself.
Make no mistake about it: There is no “debate” still happening here. The debate is over, and Tesla won. Everyone with a brain knows that LIDAR is a crutch.
Thurn defects diplomatically suggesting “he’s glad there are different hypothesis”, but it seems pretty clear he was just throwing a bone to his former employer and colleagues. After an entire decade with little progress, I predict 2020 will be the decade everyone realizes Waymo is a massive failure.
The founder of Google’s Self-Driving Car Project (Waymo) has a Tesla. Here’s what he had to say about Autopilot:
Tweet
I think you might be reading a bit too much into his remark on LIDAR… My reading is that while he thinks cameras *should* be good enough, it’s impossible to tell for sure which approach is better just by musing about it — we won’t really know until both approaches have been tried, and we see which actually gets there first…
he drives a Tesla every day and says it makes him safer. Google company line is that that is unsafe and only full autonomy is safe and you need LIDAR. Him buying and using a Tesla directly contradicts Google’s official story
Sticking with LIDAR and refusing partial automation are entirely different things…
of course. did i say they are the same thing?
Uh… Sorta? You are saying that he contradicted Google’s official story on partial autonomy and LIDAR, as if it was all one story — while I’m saying that he only outright contradicted the story on partial autonomy, while he seemed rather agnostic about LIDAR…
the company’s official PR line about autopilot is they tried that approach but everyone fell asleep. clearly their early engineers disagree with the official PR line. definitely funny