Thursday, July 27, 2017

The Tesla Model 3 -- two inflection points

"It's actually even more important to design the factory than it is to design the product itself."
"Expect Level 5 autonomy in "about two years."
(Tesla CEO Elon Musk, video interview below)

Telsa will begin deliveries of their Model 3 tomorrow and I think that might mark two inflection points -- one in the maunufacture of electric-powered cars and the other in their autonomous control. To put the Model 3 in context, consider two earlier automotive inflection points, the Models T and A Fords.

Model T runabout
The first mass-market car was the Model T Ford, which began production in 1908. Ford was able to produce large numbers of Model Ts and sell them at a relatively low price because they were mass produced on an assembly line, which reduced cost and increased production rate significantly. Ford sold 10,666 Model Ts in 1909. The runabout (roadster) sold for $825 and the four-seat touring car was $850. Over the years, they refined the design and added sedan and coupe models, but by 1927 sales were falling and competitors were offering new features. Ford stopped Model T production and retooled to produce the Model A.

Model A roadster
The Model A began production October 20, 1927, and went on sale December 2. (They called it the "1928" Model A). The Model A offered more models than the Model T and a choice of colors. The mechanical design improved in many ways over the Model T and the driver controls were similar to those of today. If you know how to drive a stick-shift car today, you would be at home in a Model A, but would have to be taught how to drive a Model T. The Model A was a mass-produced, modern car and Ford had plants in Argentina, Canada, Denmark, France, Germany, Japan and the United Kingdom as well as the US.

So, why do I think the Model 3 might be as significant as the Models T and A?

Manufacturing strategy and scale

Gigafactory capacity projection (source)
Ford pioneered the assembly line in a vertically integrated factory and Tesla also hopes to transform the manufacturing process. The size and design of their Gigafactory is unprecedented.

When complete, Gigafactory 1 will be 6 million square feet, making it the biggest building in the world by footprint and second only in volume to the Boeing factory in Washington state. It will produce batteries for homes, vehicles and power companies as well as the Model 3 powertrain -- the motor and gearbox assembly. Final assembly of Model 3s will be done in a separate automotive plant, but they expect subsequent Gigafactories to incorporate the entire auto assembly.

Tesla views a factory as a machine for making machines and they have approached its design as one would approach the design of a multilayer chip. They hope to be able to improve the factory "clock speed" and "density" over time, leading to a 30% cost savings compared to other battery factories. Tesla's manufacturing strategy was outlined in this presentation at the introduction of Gigafactory 1:


Gigafactory 1 output will ramp up as it is built out and, by 2020, they expect its output to exceed 2013 global battery production. They plan to announce three or four more Gigafactories this year, one of which will probably be in China.

Tesla's "wall of patents"before
and after (image source)
Tesla's intellectual property policy is also innovative. On June 12th, 2014 they released their 249 patents, saying they would not sue anyone for using their technology in "good faith." As shown here, they took down the plaques on their "wall of patents" after releasing them, replacing them with an image and the slogan "OEMS all our patent are belong to you." (I think Yoda wrote that for them). It seems that Elon Musk sees other car and battery manufacturers as collaborators in the effort to rapidly achieve conversion to sustainable energy -- he realizes he cannot do it by himself.

Autonomous Control

Many auto manufacturers are working toward self-driving cars, but Tesla seems to be leading the pack. Elon Musk outlined the planned roadmap for the Model 3 in the following interview:


(If you are in a hurry, the discussion of Tesla cars and trucks begins about 11 minutes 10 seconds into the interview, but I'd recommend listening to the entire interview, which also covers boring tunnels, solar roofs, SpaceX and Mars).

Level 5 autonomy
Musk says the Model 3 will come with sensor hardware that will enable them to achieve level 5 automation, as defined by the Society of Automotive Engineers (SAE): "the full-time performance by an automated driving system of all aspects of the dynamic driving task under all roadway and environmental conditions that can be managed by a human driver." (Follow this link or check this graphic for the definition of SAE's autonomy levels).

Those sensors will enable an autonomous cross-country drive this year. Musk says "November or December of this year, we should be able to go all the way from a parking lot in California to a parking lot in New York, no controls touched at any point during the entire journey." He added that it could have been any two cities on the highway system in a given country "We could change it and make it Seattle-Florida, that day, in real time. So you were going from LA to New York; now go from LA to Toronto." (I wonder if it would work in nations where they drive on the left side of the street).

But that demonstration drive pales in ambition compared to his prediction that, barring regulatory constraints, they will release a software update that brings the Model 3 up to Level 5 autonomy in "about two years."

Other manufacturers are less optimistic -- why might Tesla have a lead over the others?

The artificial neural net technology employed in autonomous vehicles depends upon software design, fast hardware, and access to relevant data. Musk has made several investments in artificial neural net companies, including Go champion DeepMind, and he says his aim is not a financial return but keeping abreast of developments in the field.

That would keep him at least even with possible automotive competitors in hardware and software design, and he has a definite data-collection lead. Tesla vehicles have been online and collecting data for several years. From their inception, Teslas were conceived of as "platforms" for downloadable control software as well as data-collection input devices.

The Model T Ford was the first mass-market car and it had little competition for some time. While the Model 3 may be innovative it, like the Model A Ford, already has competition like the relatively low-cost Chevrolet Bolt and virtually all auto manufacturers are introducing electric or hybrid vehicles. (Chinese owned Volvo has announced that all the models it introduces starting in 2019 will be either hybrids or fully electric).

Of course, Elon Musk predicting radically improved manufacturing or Level 5 autonomy in two years, does not guarantee he will achieve those goals, but do you want to bet against someone who has made landing used rockets on barges at sea a somewhat routine event?

Monday, July 03, 2017

The long memory of the Internet -- Trump then and now

The Internet has a long memory -- check for yourself by googling "early Trump interviews" and filtering for videos.

In the early days of the Intenet, we naively expected its political impact to be rosy -- leading to informed, intelligent discussion and a flowering of democracy. Many of us held on to that vision as we watched the use of the Internet during the "Arab Spring," but our optimism has eroded steadily since that time. Terrorist recruiting, fake news and lying politicians have dominated recent discussion of the political impact of the Internet, but I have some good news -- the Internet has a long, albeit imperfect, memory.

This was driven home for me by a recent segment on John Oliver's TV show Last Week Tonight. After Donald Trump fired FBI Directory James Comey, he tweeted that he might have recordings of their three previous meetings. Oliver showed and commented on a Fox News interview of Trump after he admitted that he had not recorded the meetings.

Watching the interview, I was amazed by the incoherence of Trump's speech and his dull expression and tone. His wife, who was standing beside, him seemed frozen. I was so impressed by his incoherence that I searched for the clip online and downloaded and transcribed it.

Oliver introduced the interview segment by stating that:
You may remember back in May Trump suggested on Twitter that he may have tapes of his conversations with deposed FBI director James Comey. Well, on Thursday, Trump finally admitted that he had no such tapes and offered this rationale for his claim.

Here is the Transcript of Trump's explanation:

Trump: Well, if I didn't tape him you'd never know what's happening when you see that the Obama administration and perhaps longer than that was doing all of this unmasking and surveillance that you read all about it and I've been reading about it for the last couple of months about the seriousness of the and the horrible situation with surveillance all over the place and you've been hearing the word unmasking, a word you probably never heard before, so you never know what's out there, but I didn't tape and I don't have any tape and I didn't tape.

(Oliver jokes)

Trump continues:When he found out that uh I you know that there may be tapes out there, whether its government tapes or anything else and who knows, I think his story may have changed. I mean you'll have to take a look at that because then he has to tell what actually took place at the events and my story didn't change, my story was always a straight story. My story was always the truth, but you have to determine for yourself whether or not his story changed, but i did not take.

Interviewer: That was a smart way to make sure he stayed honest in his hearings.

Trump: Well, uh, it wasn't uh it wasn't very stupid I can tell you that.

(Oliver jokes)
You can see interview (2:47) along with Oliver's commentary here:


Trump's self-defeating incoherence led me to wonder if he might be mentally impaired, so I searched for other examples online and it turns out that Trump's speech patterns today are strikingly different than when he was younger. For example this survey article compares clips of Trump's earlier interview responses with those of today. Experts interviewed for the article agree that Trump's speech has deteriorated, but all qualified their observations by pointing out that one could not determine the cause without clinical examination -- it could be the onset of dementia, but it could also be explained by normal healthy aging, being tired, stress and pressure, or it might even be a strategic appeal to relatively uneducated voters. I'd throw in narcissism and obsession with Obama as well.

Regardless, the Internet has a long memory -- check for yourself by googling "early Trump interviews" and filtering for videos.

-----
update 7/4/2017

An article in the Atlantic Monthly posits another possible reason for Trump's mental decline, citing research showing that power can lead to a leader's loss of mental capacity -- a phenomenon one researcher refers to as "hubris syndrome."

I experienced this personally when I spent a year and a half as a consultant to the CEO of a large corporation. I was in many meetings with the CEO and various managers and vendors. People jockeyed to sit next to him around a conference table and seldom disagreed with anything he said. It was a status symbol to refer to him by his first name. I had the strong impression that being in a status bubble all day for years had made him somewhat narcissistic and overconfident.

Similarly, Trump is the boss in business and a fan of his cheering, enthusiastic base at political rallies. Perhaps he cannot conceive of being wrong, resulting in flustered incoherence when he is criticized or asked a probing question. Few people would have a sufficiently strong character, sense of purpose or justice not to be affected by being surrounded by "yes people" for years.