At an autonomous driving event for investors at Tesla headquarters in Palo Alto, CEO Elon Musk along with some Tesla engineers, showed off the hardware, the
software and the network that the company says will lead to fully autonomous vehicles by the middle of 2020.
Although the company may not have regulatory approval for complete autonomy by then, Musk said that — as far as the hardware and software are concerned, you could safely operate the car next year without any human interaction.
He even predicted that Tesla owners would be able to add their car to a fleet of “robo-taxis” managed by Tesla’s network to earn income from their cars when they might otherwise be idle. The company recently started leasing Model 3s but will not let customers buy their cars at the end of the lease because, Musk said, “Tesla wants them back,” to use as company-owned robo-taxis.
He said that Tesla vehicles currently being built are equipped with a new Full Self Driving (FSD) chip that is many times faster than the current technology and able to make instantaneous decisions in city traffic. Older cars, built after 2016, can be upgraded with the new chip, which the company says it will provide at no additional charge to those who paid between $2,000 and $5,000 for FSD software. The price anyone paid or will pay depends on when they ordered.
Musk said that “Any part of this (self-driving computer) could fail and the car will keep driving” and “the probability of this computer failing is substantially lower than someone losing consciousness.”
“Three-ton death machines”
Last month, Tesla updated its Enhanced Autopilot navigation so that is now allows cars to operate semi-autonomously on highways, changing lanes to overtake slower cars and exiting the highway or transitioning to a different road. However, drivers are required to hold the steering wheel as a safety measure. Musk predicts that it will soon be OK for drivers to sleep while they’re cars are in motion and that the company will eventually remove the steering wheel and pedals. Even current cars could be modified in the future to have the steering wheels removed and replaced by a cap, Musk said.
Musk predicted that “Consumers will demand in the future that people are not allowed to drive these three-ton death machines.” He was only slightly off. In 2018, the average car weighed 2 tons, or 4,000 pounds,
Neural networks and data from ‘the fleet’
Tesla is employing neural networks that are capable of learning over time not only how humans currently drive but how to drive better than humans. One of Tesla’s big advantages over its competitors is its “fleet,” the nearly 500,000 Tesla vehicles now on the road that are equipped with self-driving hardware, along with cameras and sensors that send data to the company. The company is able to analyze not only how the current autopilot software is functioning, but how, when and why humans override the autonomous systems. It also analyzes human driving behavior, even on vehicles that don’t have the autopilot software. All that information is used to train their automated systems to be more accurate and safer.
For example, autonomous cars would, of course, try to avoid hitting bicycles. But what about bikes that are attached to cars via a bike rack. The software would have to understand the difference. It would also have to understand the difference between a plastic bag on a road vs an object that could cause damage if run over. It needs to function on wet roads, in heavy fog, snow and other conditions along with roads with poorly marked lines or signs. But these conditions already exist in the real world, so it’s now getting the data it needs to make these judgments.
Predicting other drivers’ behavior
Have you ever made an educated guess of what a driver was about to do based on the way they are moving slightly in one direction? Tesla is working on software that can do that perhaps better than humans. For example, when someone in another car changes lanes in front of you, it can examine what that car was doing 3 seconds earlier, to develop the basis for predicting behavior to figure out what a car is going to do, Musk said.
Musk makes some compelling arguments about the value of his company’s data when it comes to creating autonomous vehicles. I live near Google’s headquarters, and it’s common to see autonomous Waymo cars driving around collecting data about roads, driver habits and other important elements that the company needs to train its software and networks. The cars are being driven around by paid employees, and, although I don’t know the size of the fleet, I’m sure it’s much smaller than Tesla’s fleet of cars owned by customers who paid Tesla for the privilege of driving them and sending data back to the company. That’s a great deal of information that Tesla can use to fine-tune its algorithms to operate cars autonomously.
Can already see progress on the way to self-driving
As a Model 3 owner who purchased Enhanced Autopilot software, I have already seen improvements in the way the car handles, even in the few months I’ve owned the car. It’s now capable of going from on-ramp to off-ramp on a highway without any driver intervention.
Autopilot is not perfect. I’ve had to take control a few times, but each time I do that, Tesla is learning what happened, how I and thousands of other drivers responded, and how it can make its software better over time.
Still, I’m skeptical about Musk’s ambitious timeline. He has made many predictions about self-driving before that haven’t panned out. And, in 2016, he said that all current cars (at that time) had full self-driving hardware only to now have to develop this new chip that will be able to retrofit cars back to that date.
And, of course, the regulatory approval process could take a long time, even if Tesla feels it’s ready to let people ride in cars without drivers.
I’m not counting on robo-taxi revenue, but I’ll take it if it comes
As per the robo-taxi promise, Musk is basically saying that Tesla owners will be able to add their car to the fleet managed by Tesla’s network. In an Uber-like strategy, Tesla will take about 30% of the revenue, leaving the rest to the owner. I’m not holding my breath for getting passive income from my Model 3, but if it actually is possible, I’d be happy to take advantage of the revenue stream when I’m not using my car which, like most people, is most of the time.
Larry Magid is a tech journalist and internet safety activist.
Published at Thu, 25 Apr 2019 00:30:16 +0000