Recently, the California DMV made public the reports it received from companies testing autonomous vehicle technology on public roads in 2016. The reports vary in the amount of data, but generally contain two key sets of facts, the amount of miles driven and the number of times that the autonomous driving mode needed to be disengaged. Tesla (NASDAQ:TSLA) was one of those filing a report, and based on this, some have concluded that Tesla is hopelessly behind in autonomous vehicle technology. Nothing could be further from the truth.
Crazy Tesla, Crazy Daimler
SA Contributor Paulo Santos is one of those who concluded that Tesla is "hopelessly behind". Santos does a good job of summarizing the various reports available on the DMV site, and for that reason alone, his article is a good read. However, I disagree with his conclusions, since I see no scientific basis for them. There simply isn't enough data available in the reports to assume that the situations being tested are comparable.
But let me leave that issue aside for a moment in order to focus on what I think is an intriguing fact. The report from Mercedes-Benz Research and Development North America and the report from Tesla disclose similar amounts of total miles driven, autonomous system disengagements, and disengagements per mile, as summarized below:
This would appear to put Daimler AG (OTCPK:DDAIF) in the same boat as Tesla in terms of being hopelessly behind the competition.
Tesla's Autopilot 2 system is based on Nividia's (NASDAQ:NVDA) Drive PX 2 hardware and software and together with an enhanced sensor suite will serve as the basis of its "Full Self-driving Capability" that is currently being marketed as a $3,000 add-on to the "Enhanced Autopilot" system. SA Contributor Enertuition has characterized Enhanced Autopilot as "immature and dangerous and certainly unfit to be on public roads".
If Enertuition is correct, then Daimler has made as huge a mistake and is as irresponsible as Tesla. As I highlighted recently, at CES, Nvidia and Daimler announced a partnership to build a self-driving car. In fact, the partnership has been working away on the problem for several years.
Furthermore, the Daimler representative Sajjad Khan announced that Daimler would have a self-driving car on the road "within 12 months", more or less the same time frame as Tesla has been promising for its Full Self-driving Capability that it announced back in October.
A Brief Anecdote
Is it possible that Daimler, one of the oldest and most prestigious automobile companies in the world, has suddenly been infected by "Tesla Fever"? Tesla Fever is characterized by a reckless willingness to market unsafe technology to unsuspecting customers. The Tesla bears often invoke some form of Tesla Fever in characterizing the short thesis for Tesla.
I seriously doubt that Daimler has been infected by any such thing. At this point, let me digress briefly with a personal anecdote by way of explanation.
I started my engineering career at Hughes Aircraft in 1984. Shortly after I was hired, the portion of Hughes that I was in, the Electro-optics and Data Systems Group (EDSG) was bought by General Motors (NYSE:GM) and I became a GM employee.
I was in the Laser Laboratory at EDSG. The laser lab designed the laser sensors built by EDSG, including laser rangefinders and laser radar, which at the time we called LADAR, although most of the time the term LIDAR is now preferred.
At the time, LIDAR capable of 3-D mapping the surroundings was very new and used mostly for military applications. I was in the group that did R&D in 3-D mapping LIDAR.
When GM bought EDSG, there was a lot of talk about technology synergies, and GM tried to find ways to apply EDSG technology to automotive applications. One of the ways GM explored was an early form of active cruise control that would detect cars ahead and slow down in order to maintain a safe following distance.
I participated briefly in a systems engineering trade study being conducted by GM to identify the best sensor option for the active cruise control system. My one and only direct contact with mainstream GM management was when I attended an outbrief of the trade study in Michigan.
There was some culture shock on my part in the encounter. I was unprepared for how smart and technically astute the GM engineers were. I was also very impressed by how seriously GM took the issues of safety and product liability. This made them inherently conservative, but in a good way. The outcome of the trade study was that GM chose a radar based system, rather than the laser based sensor EDSG was proposing.
GM's corporate culture was probably fairly representative of the auto industry as a whole, and I'm sure Daimler is right in there. If anything, Daimler is probably more conservative. While it may seem plausible to attribute irresponsible conduct to Tesla, it's not for Daimler. I don't think it's possible that Daimler would have made its announcement at CES unless it was highly confident that the system would work.
So how to explain the DMV data? There are a variety of ways. Testing may have been done outside of California, or in closed, non-public areas, all of which would be exempt from reporting to the DMV. We also know from Tesla that the Nvidia system has the ability to operate in "shadow mode" in which data is collected and algorithms tested, but there is no actual control of the car by the self-driving hardware/software. Shadow mode testing would be exempt from reporting as well.
Tesla's (and Daimler's) Powerful Advantage
The availability of shadow mode in Nvidia's hardware provides a powerful way to rack up autonomous driving miles. Since every Tesla produced will have the necessary hardware, Tesla can deploy its self-driving software to these vehicles in shadow mode. A fleet of 100,000 Tesla vehicles, driving 10,000 miles in the first year, can accumulate a billion miles of shadow mode testing in the first year.
Far from being hopelessly behind, Tesla can easily overtake Alphabet's (NASDAQ:GOOG) (NASDAQ:GOOGL) Waymo, which according to its DMV report, has accumulated a total of 2.3 million miles. As Tesla deploys features for Enhanced Autopilot and Full Self-driving to it fleet, it will also quickly overtake Waymo in actual self-driving miles driven. Daimler will enjoy a similar advantage for its self-driving system.
My professional opinion is that the Nvidia system represents the best approach and the one that is likely to achieve governmental acceptance first. This is not to minimize Waymo's achievement. It can still be argued that Waymo represents the most mature approach currently. However, Waymo is limited by the size of its testing fleet, whereas Daimler and Tesla can test (and train the AI algorithms) on a much larger fleet, thereby closing the maturity gap in short order.
One additional note on the issue of using LIDAR. Santos makes a point of Waymo's use of LIDAR as an advantage compared to the Tesla/Nvidia system which uses high definition video cameras combined with a forward looking radar. As someone with experience with the technology, I doubt that LIDAR confers an advantage.
Waymo's LIDAR system, which for a long time has been based on the Velodyne HDL-64E, shown below, does offer very good range accuracy, spec'd to be less than 2 cm.
However, it's questionable that such accuracy is really needed for a self-driving car. After all, human drivers don't measure ranges to that accuracy. It's possible to construct a 3-D map of the surroundings with HD video cameras and extensive processing. This is the approach taken with the Nvidia system.
Generally, replacing a laser sensor with processing combined with passive sensing will win out economically. For those of us in LIDAR development, this was a frustrating reality that we encountered time and again. Processors and software are relatively cheap. The Velodyne system costs $75,000. Waymo is reportedly deploying a less expensive $7,500 system, but that's still much more than the cost of a Drive PX 2.
Also, LIDAR is very unforgiving of weather conditions. A little light rain or fog will be detected as volumetric noise, more or less hosing the 3-D mapping process. Once again, video cameras are much less sensitive, and algorithms can still recognize essential features even with the "noise" of rain or fog, just as our brains do.
Although Santos and Enertuition have presented reasonable counter arguments, I continue to regard the Nvidia based systems for Tesla and Daimler as having the best chance to first achieve self-driving capability. The fact that Daimler has committed to the Nvidia system says that they think so too.
The flawed interpretations of the DMV reports being offered by the Tesla bears are yet another example of second guessing of Tesla's (and Nvidia's, and Daimler's) engineering expertise, which I consider to be an unproductive form of criticism. Especially when it comes from those without expertise in the technology.
I don't regard Tesla as above criticism, but I continue to regard Tesla and Nvidia as highly competent technically. I continue to maintain a hold rating on Tesla, which I may upgrade to a buy pending Tesla's Q4 earnings report on February 22. I maintain a buy rating on Nvidia, based in part on its exciting opportunity in self-driving cars.
Disclosure: I am/we are long NVDA.
I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.
Editor's Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.