Hacker News Reader

Comment by mkoryak | Hacker News Reader

I don't understand the current state of self driving cars.

First we have tesla which already has a self driving car. They were the first to build one and now everyone is playing catch up, or are they?

Waymo has been building something for a really long time too but I have no idea how close they are to an actual product.

Now there is this new company that gets a bunch of money to build something.

It feels to me like the self driving car market will go to the first player who can make a self driving car -- which is tesla.

Or maybe no one is close to making a really good self driving car yet, and even if they did, they would have to convince people like me that I can sleep at the wheel while surrounded by massholes.

View on HN

As someone who worked in the industry, there are a couple of different approaches, incredible amounts of hype, and disappointing amounts of progress to fully autonomous vehicles.

* Tesla driver assistance features that they market as self-driving features. You need a driver that is alert and monitoring the vehicle because it can return control to the driver without warning.

* Waymo has launched limited L4 pilot in the "easy" Tempe, AZ area. They have remote teleoperation / monitoring but no driving in the car.

* ATG has prototypes that run with safety drivers

* Cruise has prototypes that run with safety drivers + they are working on tele-monitoring / teleoperation.

* Comma.AI has a consumer-available L2 driver-assistance device that you can add to your car to do ACC and lane keeping.

* Aurora has prototypes that run with safety drivers

* Zoox (bought by AMZN) has prototypes that run with safety drivers

* Smattering of other companies that are working on the tech, but are no where close to a product launch (Lyft, Toyota, Argo etc)

In general, there was a bunch of hype and bunch of money poured into the industry in 2016-2018. Others saw people pouring money into it and assumed a breakthrough was close and so they poured more money in. The cycle continued to spiral into huge multi-billion dollar investments into technology that is still IMO quite far away from a public launch. Turns out that several years later, it's really hard and no one is actually that close to useful AVs.

Waymo has a publicly available product launched, but I'm confident they are running that program at a significant loss.

Cruise has struggled to hit the milestones set in their funding deals to unlock the next levels of funding.

In my opinion, Comma is going to crush it in the next 2 years, Tesla will keep refining their AP tech, but still won't be FSD. I think autonomous robotaxis will be the norm eventually but not until after 2025.

> it's really hard and no one is actually that close to useful AVs

My suspicion is that the thing everyone's going to founder on is that a large part of safely operating a motor vehicle on the public roadways consists of observing and then predicting the behavior of all the drivers, cyclists and pedestrians. So you need to either have true AI, or somehow get all the humans on those roadways to stop behaving so erratically. The problem is, one of those is currently impossible and the other is currently infeasible.

(I'm not sure which is which, and I'm not sure it matters.)

Are you concerned that lack of progress could lead to a sort of "self-driving winter," like the AI winters experienced in the past?

> In my opinion, Comma is going to crush it in the next 2 years, Tesla will keep refining their AP tech, but still won't be FSD. I think autonomous robotaxis will be the norm eventually but not until after 2025.

Curious why you think Comma is going to crush it vs all the other companies working in this space?

I think Tesla and Comma are by far the best positioned for the consumer tech. Trucking is not something I feel qualified to comment on, because I don’t have a sense for the economics. But I know a bunch of R&D is going on here.

Most players in the space made investments in driver replacement tech which is multiple years from profitability. I think the places that this kind of tech makes sense is in licensing it to TNCs or operating a fleet of vehicles that you list on a TNC’s network.

On the other hand, driver assistance features are the only systems that can generate meaningful revenue today. One of Tesla’s finest selling points is their AP software.

In this vein, Comma is the Android to Tesla’s tightly integrated iOS. They are profitable today and ultimately have a larger TAM (new cars vs Tesla owners).

> First we have tesla which already has a self driving car. They were the first to build one and now everyone is playing catch up, or are they?

They have a Level 2 system, which is certainly an impressive technical achievement, but not what most of us are thinking of when we think, "self driving." They are hoping to grow it into something more, but it remains to be see whether that will be possible. Other companies also have Level 2 systems, and, while I've personally never used any of them, some sources do report that Autopilot is not the best one out there, just the most hyped one. So, even if we are just focusing on assisted driving rather than self driving, I'm not sure one can say with too much confidence that they are the clear leader.

I would not interpret the existence of a functional consumer product as evidence of a technical lead in the race toward true self driving. Waymo never had any intention of commercially releasing a level 2 system. Their plan was to go straight to level 4 without any detours. Reports would seem to indicate that they are far, far ahead of everyone else on achieving that goal. It may be that they now have an unbeatable lead. Or it may be that some more direct competitor catches up. Or it may be that a company like Tesla or Cadillac is able to use profits and data collected from their commercial level 2 systems to accelerate their development and catch up. This, too, remains to be seen.

My intuition is that Waymo and Tesla dominate in the amount of data available to them.

Waymo seems to be closest to a robust 3D lidar based solution, but Tesla maybe be closest to 'a' solution, but the community is split on whether their 2D approach will work.

Then there seem to be a laundry list of companies that entered the market late, and are propped up panicking car companies which aren't sure which one of these companies is the real thing. So, they hedge their bets by taking a piece of all of them.

The solution will come out of 1 of the following:

1. A slow race to making a more and more robust vision system that is eventually good enough to get through regulation. In this case, the earliest to start and one with the most data will win. So, either Waymo or Tesla. By the time others get there, they will have captured the market and made billions.

2. Some smaller company has an 'a-ha' moment. Waiting for an a-ha moment doesn't sound like the most prudent investment decision, but the one that finds it will become a billionaire overnight. This change is so important (like Mapreduce) that it will quickly be reverse engineered and one of Google/Uber/etc. will release a self driving hardware kit + api or come pre-installed in cars like android auto. The solution transforms the vision field with a reach far beyond just cars.

Put down the kool aid. There are no self-driving cars.

To be fair, the levels of self-driving are almost never discussed so it's no wonder people say things like "Tesla already has self-driving cars".

For those unaware:

- Level 0: No automation

- Level 1: Driver assistance (Assisting steering or braking)

- Level 2: Partial automation (Assisting steering AND braking)

- Level 3: Conditional automation (Human supervised autonomy)

- Level 4: High automation (Full automation in ideal environments)

- Level 5: Full automation

I believe we're currently at Level 2-3.

More details here: https://www.howtogeek.com/401759/WHAT-ARE-THE-DIFFERENT-SELF...

If someone with more knowledge wants to correct me on the levels, feel free.

There's nothing "to be fair" about this.

The levels are technical distinctions in a roadmap towards "self driving cars", which by any reasonable definition are Level 4 and 5.

Below that it is marketing for advanced cruise control features (level 1 and 2), or dangerous (level 3).

I own a Tesla with FSD. Can confirm, there is no self driving car.

To be fair, I started with "I don't understand the current state of self driving cars".

> First we have tesla which already has a self driving car.

What 'self driving car'? I think you meant 'semi-autonomous cruise control' as there is no Tesla car that is fully 'self driving'.

As for Uber's situation, they had no chance. Mounted losses everywhere caused by the pandemic, government lockdowns and drivers who are now classified as Uber employees make the costs of it all unsustainable for them to justify keeping a 'self-driving unit'.

The only certain and sustainable long term players in this 'self-driving' or 'cruise control' game are Tesla and Comma.ai.

I was fascinated to learn that Comma are the only (?) company whose ML algo dynamically segments the road and surroundings, whereas I believe Tesla specifically looks for things like lines on the road.

[https://medium.com/unikie/self-driving-cars-2020-and-beyond-...] helped me understand the major players in self driving vehicles and what they're upto.

Tesla does not use Lidar as a cost savings measure, and only uses camera output to determine how the car drives. It is unknown if this is enough to account for enough permutations of traffic, but they're calling it full self driving while using car owners as paying (!) beta testers.

AFAIK, no one is close to releasing an autonomous vehicle that can drive on par or better than an average human driver, but that several have a path to getting there.

I believe many companies can thrive until M&A whittles down that list. I don't think this is a first mover takes all kinda deal because safety will have to be proven over a long time allowing other entrants to prove their worth.

>Tesla does not use Lidar as a cost savings measure

I think it is less about cost saving and more about getting to the root of the problem. According to Elon's statements, Lidar is a crutch and since the AI will eventually have to be good enough to remove the crutch, remove it an the get go and save learning how to use the crutch. This is similar to his reasoning about EVs. Since we will eventually have to have petroleum free cars, Lets remove them as early as possible.

Sometimes you need a crutch because your legs are not up to the task of walking. And the only way to recovery is to use them

> According to Elon's statements, Lidar is a crutch and since the AI will eventually have to be good enough to remove the crutch, remove it an the get go and save learning how to use the crutch.

There's zero reason to believe him on this.

LIDAR - combined with vision - gives a much better awareness. The only reason not to use it is (or was) cost.

Its a cost savings measure. Tesla is adding an imaging radar to their cars (which gives a LIDAR-like output) because their current sensor suite doesn't cut it. "LIDAR is a crutch" is Elon trying to rationalize why he advertised cars without the necessary hardware as capable of autonomy.

(See https://mobile.twitter.com/greentheonly/status/1319363515610...)

> First we have tesla which already has a self driving car

No, they don't. And it will take a while to get there.

I think there are multiple companies that can make something that works most of the time in good conditions, but you have to have a certain level of arrogance and self confidence to be the first to release it as a product.

Reply on news.ycombinator.com