Creepy as it may be, it's the best bet we have at safe autonomous vehicles.
Part of the genius of Elon Musk’s venture into the notoriously difficult to crack auto market is that, using different lenses, one can see how effective his multifaceted approach is to advancing the automobile. Electric powertrain aside, it’s the array of cameras and sensors that are loaded into every new Model S and X that place Tesla way above the autonomy playing field when compared to its competitors. As Bloomberg Technology reports, while some may look at a Tesla and see a semi-autonomous vehicle, Musk sees it as a data collection device.
The accomplishment of Autopilot is that, while companies like Google and Apple spend billions of dollars testing self-driving cars, Tesla has managed to capitalize on the public’s interest in self-driving vehicles and attract a customer base willing to pay to be a test driver. Besides, there’s no better type of test than a real world test, so Musk's approach of satisfying curiosity while gathering data sounds like a win win. In Tesla’s recent third quarter report, the company stated that it had amassed over 1.3 billion miles of Autopilot data, far more than any of its closest competitors. This statement is in contrast to what Musk tweeted in early October when he said, “Cumulative Tesla Autopilot miles now at 222 million.”
However, the number cited in the Q3 report takes into account that Tesla’s Autopilot has a “shadow mode” that peeps at what the sensors are seeing even when the system is shut off. This effectively turns each Model S and Model X into a data collection device for Tesla that navigates the real world with an unpaid (and therefore more likely to make real-world mistakes for the car to learn from) test driver at the helm. No company that’s working on a self-driving car wants to give away how far it is from full autonomy, but, according to Nidhi Kalra, a senior information scientist at the Rand Corporation, Tesla is in the lead. “There’s no question that Tesla has an advantage,” said Kalra.
“They can learn from a wider range of experiences and at a much faster rate than a company that is testing with trained drivers and employees behind the wheel.” As an expert in the field who has co-authored a Rand report on autonomous cars, Kalra believes that driverless cars will need to collect between a hundred million to a few hundred billion miles before they are more trustworthy than human drivers. Tesla is fast approaching that mark, and with the hardware of Autopilot 2.0 being added to each new model that rolls out of the Fremont, California factory, the company’s data collection techniques are only poised to get better.