And perhaps more importantly: what happens to the money customers have already paid for it?
It's been two years since Tesla started selling the full self-driving option. Was Elon Musk too optimistic? Or did he know that his believers were willing to go down his yellow-brick roadmap?
A week ago, Tesla quietly pulled its full self-driving (FSD) option from Model 3, Model S, and Model X. Tesla announced, or more accurately, Elon Musk tweeted on Oct. 18 that FSD is currently off the menu because it’s “causing too much confusion.”
This is no small matter considering that for two years, Tesla has been charging customers $3,000 to $5,000 for a so-called FSD upgrade. Tesla has not committed to refunding any of its FSD windfall.
Call me old-fashioned, but this is the sort of corporate behavior, in my book, that appears in the “scams” chapter.
Until recently, Tesla’s original order page promised: “All you will need to do is get in and tell your car where to go.” It added: “Your Tesla will figure out the optimal route, navigate urban streets (even without lane markings), manage complex intersections with traffic lights, stop signs, and roundabouts, and handle densely packed freeways with cars moving at high speed.”
Seriously, though, as Phil Magney, founder and principal at VSI Labs, noted, “It’s been two years since Tesla started selling the FSD option, and buyers of the upgrade have nothing to show for the thousands of dollars they spent. Musk implied that FSD was a year or two away, but Tesla doesn’t seem to be much closer to FSD now than they were two years ago.”
The question, then, is whether “Musk was too optimistic about how quickly full self-driving capabilities could be achieved,” noted Magney. This interpretation strikes me as exceedingly charitable.
However, I realize that Tesla is deemed by many in the tech industry to be a whole different kind of carmaker, with a fan base that comprises a whole different kind of buyer. It’s possible that Musk knew all along that his believers, dazzled by his vision of the “future,” were willing to go down his yellow-brick roadmap even if the trek proved to be two years long — or more.
An apt analogy in recent media coverage comes from a Tesla customer comparing the FSD option to a crowdfunding campaign. By backing Musk’s project, customers are excited about the concept behind the project and want it to succeed. They understand and accept that the project has risks and that the product may be delivered late — or not at all.
All of the car OEMs on the planet must be wishing that they had Tesla’s luck.
I’ve also noticed that Tesla commands a lot of respect from the engineering crowd. One engineering executive who works in the automotive space, speaking on the condition of anonymity, told me, “Tesla is no big car OEM.” That means that Tesla, unlike the legacy carmakers, does not have the luxury of leveraging two different platforms — one for ADAS and another for fully autonomous vehicles — to perfect ADAS and FSD in parallel. Collecting several thousand dollars for the full self-driving upgrade must have been Tesla’s business decision, he theorized, “to spend that financial resource to focus on R&D for their future vehicles.”
Strangely enough, that logic sort of made sense to me.
However, putting my consumer hat on, I could’t help but wonder why we’re letting Musk get away with such an obvious bait-and-switch.
Here’s the thing: Has anyone among EE Times readers in the technology business ever believed that any Tesla vehicle has the horsepower (processing power), redundancy, and sensors to make FSD possible via over-the-air software updates?
Mike Demler, senior analyst at the Linley Group, told us, “I’ve been a skeptic since [Musk] started making that claim back in 2016.”
Demler said, “The confusion he is claiming as the reason for pulling FSD can only be his own misunderstanding of what FSD means.”
He noted, “It should have been obvious to anyone looking at the technology in his cars that there’s no way they could be ‘fully autonomous.’ For that matter, there are still no vehicles capable of full (i.e., Level 5) autonomy. His blog came a few months after the first autopilot fatality, when he began replacing Mobileye EyeQ 3 systems with Nvidia hardware. Now he’s planning to replace that with a Tesla-designed SoC.”
Demler added, “The earlier systems weren’t designed for anything beyond Level 2, so it was a totally unbelievable claim. I doubt their new chip will enable anything beyond a limited set of Level 3 features, as in Audi Traffic Jam Assist and Cadillac Super Cruise.”
Have we nailed ADAS yet?
We already know that the processing power necessary for autonomous vehicles is huge. But the problem doesn’t end there. Ask any sensor technology supplier. You’d be hard-pressed to find a vendor who could state flatly that the automotive industry has even nailed ADAS yet, let alone fully self-driving cars.
Chris Jacobs, vice president of autonomous transportation and automotive safety at Analog Devices Inc. (ADI), acknowledged that barriers to full autonomy are still very high. Among the biggest are an obvious lack of legislation for highly automated vehicles (HAVs) and insurance policies associated with them. Neither issue is close to resolution, he said. But equally important are sensors with adequate high resolution and better algorithms.
“Today’s ADAS system can function as a warning system,” said Jacobs. “But when it comes to Level 3 actuation, we still have a long way to go.”
Many radars in use today are still primitive in the range and resolution that they offer.
Jacobs holds high hopes for a new generation of “imaging radars,” which he called “utility players.” He expects them to play a critical role, but those high-resolution imaging radars won’t become broadly available until 2020 model cars (Level 2 and Level 3 cars) and Level 4 robo-taxis, he explained.
Asked about lidars, he said that they “still have a long way to go” before attaining “automotive-grade performance, price, and ruggedness.” However, he believes that lidars will play a unique role. Picture yourself driving on the Autobahn at 200 miles per hour at night. Suddenly, you see a truck tire in the middle of the road. In minimal reflectivity and range, an autonomous vehicle needs to rely on lidars to spot the tire. Cameras and radars can’t do it, said Jacobs. “It may be 2025 or beyond for a lidar to exercise its strength,” he noted, but lidars are a must for fully autonomous vehicles.
Another critical sensing element is an “inertial measurement unit” (IMU). Calling IMU “an unsung hero,” Jacobs said that high-performance IMU is crucial for dead-reckoning applications and navigation. IMU stabilizes aircraft and missiles. “Think about an autonomous vehicle moving into a tunnel.” Jacobs explained that cameras, suddenly in different lighting conditions, will be unstable. Radar waves will start bouncing around the tunnel walls. To prevent a self-driving car from drifting out of its lane, IMU is a crucial last resort, he noted.
Musk famously claimed that a self-driving vehicle needs no lidars. Most experts disagree. Demler told us, “I have a picture of a Tesla with manufacturer plates, which I took when it just happened to pull up next to me at a light in Palo Alto. It had a Velodyne lidar on the roof. So [Musk] is not as dead-set against that as some of his statements imply.”
Class action suits coming?
VSI Labs’ Magney noted, “Tesla has not commented if they will return any of the money they’ve collected for FSD these past two years. Tesla still plans to work on FSD and roll it out to customers via OTA at some point in the future for those people willing to pay for it.”
That may be so, but Tesla has backed away from predicting the arrival of FSD.
Magney suspects, “Perhaps Tesla was worried about a class action lawsuit from FSD buyers or an inquiry from the government looking into selling a future service that never materializes.” He added, “Whatever Tesla’s motivation, dropping FSD as an option is a clear indication that Tesla won’t be able to deliver it anytime soon.”
The consensus among automotive experts is that Tesla won’t be able to do full self-driving updates until it updates its hardware. While nobody doubts Tesla’s ability to eventually deliver FSD, nobody is venturing a realistic estimate of how much it will cost.
Demler added, “With all the negative publicity the autopilot failures have generated, I hope Musk had some advisers telling him to stop claiming FSD capability. Perhaps they also said that we should wait until we have our own chip, but the decision could be completely unrelated. There have been several lawsuits over false autopilot claims, some that Tesla just settled.”
Make no mistake, though: Nobody is saying that Tesla is mothballing FSD.
Magney said, despite all the news, “Tesla is not holding back updating the capabilities of Enhanced Autopilot features.” He explained that Tesla just released SW version 9, “which actually happened sooner that I would have expected.” According to Magney, “You now have L3 capabilities even though, technically, it is not L3 per the SAE definition as the driver must stay in the loop.”
New algorithms and enhanced safety features come along with this [SW version 9] update. “For example, Tesla now offers proper blind spot detection.” Magney concluded, “I don’t expect Tesla to slow down on their distribution of Autopilot updates … about every six weeks or so. Version 9 was by far the most significant update to Autopilot.”
Nobody was able to tell us definitively the direct relationship between Tesla’s new SoC and FSD. While Magney suspects that the SoC (AI accelerator) will likely be necessary to support FSD, he said that FSD probably won’t be ready for the release of the new hardware 3.0.
Magney added, “I also believe (and there is some evidence of this) that Tesla will include map-based localization features to improve the safety and performance of both enhanced self-driving and full self-driving.” He said that this is necessary to support proper localization, lane change, and interchange support. “And most importantly, using precision maps will permit some of the advance features, such as Drive-on-Nav, which come with AP version 9.”
Meanwhile, Demler said, “I have seen more rational forecasts for when Level 4 and Level 5 might become a reality, but that hasn’t slowed down any R&D efforts [for fully autonomous vehicles], as far as I can tell.”
He noted, “The most significant trend lately is all the partnerships forming to develop autonomous-vehicle technology, such as Honda-GM Cruise, Toyota-Uber, etc.” That, however, “shows you how difficult it will be for Tesla to go it alone.”
Even though Tesla’s FSD option may no longer exist on the order page, Demler found that one can still read the same language on Tesla’s website:
We are excited to announce that, as of today, all Tesla vehicles produced in our factory — including Model 3 — will have the hardware needed for full self-driving capability at a safety level substantially greater than that of a human driver. Eight surround cameras provide 360-degree visibility around the car at up to 250 meters of range. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength, capable of seeing through heavy rain, fog, dust, and even the car ahead.
Time for Tesla to quietly remove this page?
— Junko Yoshida, Global Co-Editor-In-Chief, AspenCore Media, Chief International Correspondent, EE Times