Blog

Waymo vs. Tesla

People sometimes simplify the differences between Waymo autonomous driving and Tesla FSD as Lidar-based solution vs. a vision-based solution, especially as Tesla has been saying it doesn’t use any lidar.

But there are several other important distinctions.

For example, the production & scale is different. Tesla owns the mfg and has been selling cars to consumers, a lot of cars. Meanwhile, Waymo deploys the solution on other carmakers’ cars. As Waymo’s fleet is much smaller than Tesla’s annual delivery, the cost structure can be very different. Tesla can enjoy better economy of scale vs. Waymo, even if the hardware is the same.

Secondly, the responsibility is different, which is a key difference and debating point for robotaxi going forward. Tesla is reluctant to take on responsibility for its FSD solutions as the cars are sold, but Waymo owns the car and operates the ride-hailing service. Waymo takes the responsibility if there is accident due to the autonomous software.
On a separate note. if you think about Uber, that’s actually is a very good business model. It can gain from the value creation of autonomous driving potentially, but because it’s just a platform, it’s just matching cars and passengers and take a cut from fares, so it doesn’t need to take responsibility for a autonomous driving software failure. Uber is not a provider of the traveling service but is just providing the matching services.

There is a third element, which I am not hundred percent sure about. It is said that Waymo relies on hard-coded rules and local data vs. Tesla currently more like a blackbox. So the programmers write specific instructions to tell Waymo cars what would do in different (extreme) scenarios. Instead, in Tesla’s current version (end-to-end), it is using a lot of data the train the AI mode without human specifying what to do in each case. Tesla is only feeding data to the AI, and let AI learn from human drivers. 
This is also why some argue that Waymo is much harder to scale.

However, I do think Waymo is underestimated –

1/ it can actually take a similar approach as Tesla as well, because Waymo also has a lot of data – maybe not as much as Tesla, but Waymo for example has a lot of data in San Francisco. Waymo probably already has the best driver (or at least very good) for San Francisco. If you have the “best driver” in San Francisco and all the related data, you can probably train an AI model with what you have – Waymo can train an AI driver w/o specific rules, but based on data from its current fleet in San Francisco, which is running w/ handwritten rules by human programmers. And from that you are also replicating what Tesla was doing – basically this AI driver for San Francisco is just a learning the best practices of how to drive in San Francisco from the existing Waymo cars. I don’t think Tesla has significant data advantage if we are just talking about San Francisco.

2/ and here comes another bold assumption, which is that if you are the best driver (AI version, not specific-rules based version) in San Francisco (plus Phoenix), you are probably not a bad driver in other cities. Of course you don’t know about the specifics about other cities and other countries, so you are not as good in NYC as you are in SF, but you also won’t be as bad as people expected. And over time maybe in just a few weeks this “San Francisco guy” can do a decent job in New York City as well. 
If that’s true, and that’s probably a big if, then Waymo’s solution can actually be more scalable than people would have expected.

What I’m trying to say is that, at current stage, I don’t think Tesla’ choice technology has already won this autonomous driving competition with huge data advantages. The jury is still out.

When was Waymo approved?

In 2017, Texas passed a bill to allow driverless cars on the road.

Later that year, Waymo started to bring driverless cars to the road in Texas.

California introduced rules around driverless testing on public roads in Feb 2018.

Waymo won the first driverless permit to test in California in Oct 2018 for ~3 dozen cars.

…

In 2020, Waymo started to open its fully driverless service to the general public in Phoenix.

…

In 2024, Waymo offered the service to anyone in SF.

Lots of small steps.

Each state/city can be different. Requirements can be different & definition of “driverless” can be different.

Area can be limited.

Target passengers can be limited.

etc.

Where was Waymo’s technology at?

In 2018, Waymo’s miles per disengagement was 11,154 miles.

In 2023, Waymo’s miles per disengagement was 17,311 miles.

On average, people may drive 10k+ miles per year in the US.

So on average you will only experience one “Disengagement” in a year in 2018, which is a decent rate.

Where is Tesla FSD at?

The latest 12.5 seems to have 1 critical disengagement per 123 miles?

This needs to iterate & improve over time to be fully driverless.

A “crowd” that works

What’s a good “crowd”?

How do people know they can trust themselves collectively?

Democracy in the US won’t work if the mass can’t think/vote reasonably or behave as “adults”.

If we just use words like “respect” each other – well it’s a good quality but not enough.

I think of some key elements:

a) a crowd needs to be able to deduce correctly – like 1+2=3. Given “1” and “2”, they know those two add up to 3.

b) a crowd needs to be able to double check the facts – like they can verify “1” is “1”, so that misinformation won’t cause much harm.

c) a crowd needs to be able to correct itself. If somehow 1+2 = 10 is the mainstream idea, the crowd can find out they were wrong and make it 1+2=3 in a timely manner.

d) in many more cases, it’s not a deterministic problem. There is no one correct answer, but probably a rang of answers like 1.5-2.5. In this case, people who say 2 don’t need to correct people who say it’s 1.5 or 2.5. Or there is no right or wrong, but just a preference between 1-10. In that case, people needs to recognize the fact that there is a distribution, and a “10” shouldn’t kick a “1” out of the crowd, or vice versa.

 

China is back with good policies. What’s next?

China’s stock market is back. Everyone know.

China rolled out policies to help its stock market & economy. This makes sense.

What are the missing parts?

Some might say geopolitical tension is still an unsolved issue.

I won’t disagree. But I feel it’s not as extreme as before.

Some may say a deeper structural reform.

Fair; but in short term this might not change.

What I think is missing (and China can start to think in short term) is how China can better participate & even organize global affairs & how the world economy can grow better with a stronger China.

Historically, China didn’t excel on this. During times when its economy was leading the world, it didn’t take much care of the world. It could give stuff to nearby countries; it could invent and other can learn; however, it didn’t actively “organize the world”.

A strong economy that can’t lead is a loss to the world, I think.

How to lead? How to make other countries live better as China is a bigger part of global GDP? Not by giving free money but in a healthy, organic way.

That’s the question I haven’t seen much discussion.

Random thought on one-child policy

China implemented one-child policy from 1980 to 2015 (36 years).

This might be the largest social experiment that has ever been done globally.

Only-child policy is rare. Only China has done this.

The impact could be negligible for some families, as they would have had only one child anyway.

But collectively, this has significant impact – e.g. it reduced the population: two ppl turned into one people.

One consequence that many might have ignored – half of the families may “lose” their “last name”.

In China, children usually keep their father’s name. So families with only one girl as their child would “suffer” from this one-child policy.

This might be in China vs. say in the US, as Chinese last names are shared among different families but last names in the US are often unique. If there were one-child policy in the US, ultimately, half of the last names would disappear!

What a loss of culture.

Uninvestable? Part of the game

Sure people can argue they were seeing many absurd things in China over the past couple of years, but in the grand scheme of things, this may also be considered a part of the game.

Just like in some countries, there are various forms of “cost of doing business”.. including developed countries. They are different. So one often needs time to study, and may feel unacceptable.

China is the same; maybe a bit more special – unpredictability of rules is a cost of doing business. Maybe somewhere else, the cost of doing business is “written” in the rules so you know the “cost” before you play. Here, it’s like a higher level of “costs” – changing rules is part of the game & thus this “cost” is hard to calculate/predict. If you can’t calculate costs, and you don’t know the probability, you can’t calculate your return.

However, when you recognize that unpredictability is part of the “game”, then it can become “investable” from time to time.

The “best” player might be able to predict when is investable and when is not.

It’s also okay to be a bit behind. The key is to protect yourself when it’s “uninvestable”, and be reasonably participating when it’s “investable”.

Focus on those periods. You may temporarily forget other “costs”. But you can only pretend that you forget, with a reminder/timer. You have to remember this cost on a deeper level.

The safest period is right after the most dangerous period.

Normally, I prefer not to play poker with Chinese, as they often too good at playing with people.

This doesn’t mean I can’t play. It’s just too energy-consuming.

Smart cars

I have had this concerns about smart cars for quite a few years.

I could be old-school on this, but I like ICE, and I like traditional cars that are not connected / fully controlled by electronics.

The recent Lebanon pager explosions is not only for pagers.

EV batteries could have more powerful explosion.

In addition, EV can move.

More importantly, they can go autonomous.

In some sense, they could become autonomous moving “bombs”, especially if hacked by malicious actors.

Of course, a tank of oil (ICE) can do the same, with some autonomous features.

So it’s really important to build trust between countries and companies.

I don’t know if that’s happening.

A story about elevators

Something I recall vividly – during college, I live in a high-rise on campus. Before we push button on an elevator, we look at what’s already be pressed. If one floor above or one floor below has been pressed, we shall not push button. Instead, we walk the stairs.

This is such a good norm that I follow immediately. This is such a powerful norm that brings positive energy to people.

This is actually faster if you run a bit, as the stair is close to elevators.

More importantly, all passengers’ time combined will be saved.

I have never experienced this in China. For myself, I can’t find stairs actually..

I hope this is not only for elevators and we may replicate this spirit in other areas.