VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance • Hacker News Board

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#41

Post by jillesvangurp » Sat, Jul 13, 2019, 9:33 AM UTC

dmix wrote: post #28

jillesvangurp wrote: post #13

Earlier quoted context omitted.

Most of the more expensive EVs come pretty close to that and seem to be competing on getting the numbers better there. E.g. I think Tesla is planning to break the 400 mile barrier pretty soon. 300 mile is pretty normal already. Most EVs charge pretty quickly to about 80%. 20 minutes to half an hour gets you plenty of range. Mostly EV owners don't charge to 100% unless they are at home or doing some overnight stop. I…

Have EV companies completely abandoned the swappable battery idea? I always thought thatd be the endgame. Some standardized battery design where you roll up on top of a station and it swaps the battery underneath the car. Or maybe this is something your car does for you before arriving. Which eliminates the amount of dead time tens of thousands cars will need to spend while using up expensive property in cities while…

I don't think that ever was a serious plan.

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#42

Post by choppaface » Sat, Jul 13, 2019, 9:43 AM UTC

amelius wrote: post #27

choppaface wrote: post #16

Earlier quoted context omitted.

Uber proved it's possible to burn a lot of money on a team that is ineffective due to needless infighting and excessive internal competition.

In other words, you can't just throw money at the problem.

There are some things that money can accelerate and some things it can't. Building a balanced, effective team is not a clear linear function of money.

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#43

Post by vladislav » Sat, Jul 13, 2019, 9:55 AM UTC

konschubert wrote: post #17

boxcardavin wrote: post #2

This is bizarre, I’m in non-car autonomy but I talk daily with car autonomy folks. Everyone has been expecting a consolidation of companies through acquisitions but drive.ai showed us that it’s probably not going to happen. At The Information mobility event last month one of the SoftBank fellas said that autonomous cars will be ‘won’ by those who can outspend everyone else. Looks like backing up billions with more bi…

> one of the SoftBank fellas said that autonomous cars will be ‘won’ by those who can outspend everyone else This may not be true if there is still a fundamental leap to be made

yeah it's called unsupervised learning

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#44

Post by choppaface » Sat, Jul 13, 2019, 9:58 AM UTC

dasbo wrote: post #31

choppaface wrote: post #14

Argo also released a relatively massive 250GB dataset recently: https://www.argoverse.org/ The problem of Perception has a long road ahead at any of these companies and Argo is taking a very long position with this funding round. The headcount growth is rather worrisome but other than Aptiv there aren't many fleets that span multiple continents.

It's best not to think in GB terms when talking about AD datasets. E.g., when you record raw data of a multisensor setup (lidars, radars, cameras), the data rate can reach 10+ TB/h. Camera-only datasets are in comparison much smaller. Taken out of the argoverse dataset description: - One dataset with 3D tracking annotations for 113 scenes - One dataset with 327,793 interesting vehicle trajectories extracted from over…

Right, the dataset is not remotely useful for commercial use. For one thing, the license prohibits such use.

Yes, it's certainly not yet clear if the dataset is large enough to capture useful variance. But unlike kitti or cityscapes, it's large enough to present a computational challenge to most of the machines & budgets in research use today, so there's a pretty good chance it'll help push the state of the art... Perhaps for more than one art. The API code itself has a lot of low-hanging fruit: https://github.com/argoai/argoverse-api

nuscenes is good too https://www.nuscenes.org/

Waymo will have one as well some day https://waymo.com/open/

One attractive aspect of these datasets is that they help open up the question of safety for public discussion. For example, now anybody can throw off-the-shelf object detection at these datasets and see what a realistic F1-score looks like for objects at 30m, 50m, 100m, etc...

In argoverse and nuscenes, you have track labels, so you can furthermore factor the velocity difference into how you weigh the error. Have you ever been hit by an Uber? Even 5mph can cause a lot of damage.

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#45

Post by Invictus0 » Sat, Jul 13, 2019, 9:59 AM UTC

jillesvangurp wrote: post #41

dmix wrote: post #28

Earlier quoted context omitted.

Have EV companies completely abandoned the swappable battery idea? I always thought thatd be the endgame. Some standardized battery design where you roll up on top of a station and it swaps the battery underneath the car. Or maybe this is something your car does for you before arriving. Which eliminates the amount of dead time tens of thousands cars will need to spend while using up expensive property in cities while…

I don't think that ever was a serious plan.

The company doing this went bankrupt.

https://en.m.wikipedia.org/wiki/Better_Place_(company)

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#46

Post by baybal2 » Sat, Jul 13, 2019, 10:05 AM UTC

Hydraulix989 wrote: post #35

baybal2 wrote: post #11

Earlier quoted context omitted.

> SoftBank fellas said that autonomous cars will be ‘won’ by those who can outspend everyone else. Listen to them more. What else do you expect fund managers to say? We will not see any real self driving in our generation, and may be not even after. That is much more for practical consideration than that of technical possibility. Second to that, just how many self driving and AI startups are plain fraud? The amount o…

> We will not see any real self driving in our generation, and may be not even after. I think we can once we build purposeful infrastructure (think cars "on rails" with mesh-coordinated intersections). > Second to that, just how many self driving and AI startups are plain fraud? The amount of companies getting 9 digit valuations with nothing more than OpenCV hello worlds should make people to at least scratch their h…

> I think we can once we build purposeful infrastructure (think cars "on rails" with mesh-coordinated intersections).

In that way, surely, but you will need to build a purpose built road network for that. Some self driving buses in China work like that.

New, growing cities with centralised planning can afford that, others not so much.

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#47

Post by mac01021 » Sat, Jul 13, 2019, 10:07 AM UTC

da_chicken wrote: post #8

ml_basics wrote: post #4

"Argo AI’s focus remains on delivering a SAE Level 4-capable SDS to be applied for ridesharing and goods delivery services in dense urban areas." Perhaps naive question from an outsider to the AV world: what does "SAE Level 4-capable SDS" mean?

http://cyberlaw.stanford.edu/blog/2013/12/sae-levels-driving... It's a fairly high level of automation. One step below total automation. That is, the system controls all aspects of driving, but there are still situations where it will ask the human driver to take over. However, even if the human doesn't actually take over, the system can still handle it. It's basically capable of dealing with an odd situation, but it…

These definitions are rather vague.

If the car is able to function even when noone responds to the request for intervention, how is that different from level five? Does it mean that it safely pulls over, where level 3 would just drop control of the vehicle in a dangerous scenario?

Also, does level 5 entail general intelligence, in order to do things like follow the instructions of a traffic cop or parking attendant?

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#48

Post by petra » Sat, Jul 13, 2019, 10:08 AM UTC

baybal2 wrote: post #11

boxcardavin wrote: post #2

This is bizarre, I’m in non-car autonomy but I talk daily with car autonomy folks. Everyone has been expecting a consolidation of companies through acquisitions but drive.ai showed us that it’s probably not going to happen. At The Information mobility event last month one of the SoftBank fellas said that autonomous cars will be ‘won’ by those who can outspend everyone else. Looks like backing up billions with more bi…

> SoftBank fellas said that autonomous cars will be ‘won’ by those who can outspend everyone else. Listen to them more. What else do you expect fund managers to say? We will not see any real self driving in our generation, and may be not even after. That is much more for practical consideration than that of technical possibility. Second to that, just how many self driving and AI startups are plain fraud? The amount o…

>> We will not see any real self driving in our generation,

I agree.

But we'll be driven a large percent of trips, in the near future.

There's on-demand shared(few people in a vehicle) riding - which can be quite cheap AND fast. ridewithvia.com seem to scale with that model pretty fast. A combination of that with strong public transport systems and regulations may beat cars.

But also, a combination of partial self-driving(highway, etc), with remote control driving, could work.

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#49

Post by mkagenius » Sat, Jul 13, 2019, 10:13 AM UTC

dasbo wrote: post #31

choppaface wrote: post #14

Argo also released a relatively massive 250GB dataset recently: https://www.argoverse.org/ The problem of Perception has a long road ahead at any of these companies and Argo is taking a very long position with this funding round. The headcount growth is rather worrisome but other than Aptiv there aren't many fleets that span multiple continents.

It's best not to think in GB terms when talking about AD datasets. E.g., when you record raw data of a multisensor setup (lidars, radars, cameras), the data rate can reach 10+ TB/h. Camera-only datasets are in comparison much smaller. Taken out of the argoverse dataset description: - One dataset with 3D tracking annotations for 113 scenes - One dataset with 327,793 interesting vehicle trajectories extracted from over…

> 1000 driving hours is ok'ish for research

Wouldn't it be better to use a simulated environment first. Maybe something using pybullet[1] and a script which maps real world to STL or OBJ files[2]

1. https://pybullet.org/wordpress/ 2. https://github.com/mkagenius/osm2maya

Re: VW invests $2.6B in self-driving startup Argo AI as part of Ford alliance

#50

Post by dasbo » Sat, Jul 13, 2019, 10:38 AM UTC

mkagenius wrote: post #49

dasbo wrote: post #31

Earlier quoted context omitted.

It's best not to think in GB terms when talking about AD datasets. E.g., when you record raw data of a multisensor setup (lidars, radars, cameras), the data rate can reach 10+ TB/h. Camera-only datasets are in comparison much smaller. Taken out of the argoverse dataset description: - One dataset with 3D tracking annotations for 113 scenes - One dataset with 327,793 interesting vehicle trajectories extracted from over…

> 1000 driving hours is ok'ish for research Wouldn't it be better to use a simulated environment first. Maybe something using pybullet[1] and a script which maps real world to STL or OBJ files[2] 1. https://pybullet.org/wordpress/ 2. https://github.com/mkagenius/osm2maya

Depends on your goals.

Today's virtual worlds are not accurate enought to allow to develop perception algorithms in simulation. In order to develop sensor fusion, you also need to simulate the output of all sensors including their specific characteristics. Apart from the model quality, there is another challenge: Simulation runtime, which is substantial (!) and - to my best knowledge - not even close to realtime.

If you want to develop driving algorithms that sit on top of the perception stack, then this becomes simpler. You can work on the object level (object being simulated cars, pedestrians, ...) and statistically model perception errors. This is a lot faster, which is e.g. important, if you want to run large-scale reinforcement learning to develop your driving strategy.

In any case, in the future, I would say that we will see a lot more simulation (don't forget, all major players build heavily on simulation - just look into the numbers on how many miles Waymo simulates every day) and potentially going down all the way to the sensing level, because it allows you to develop and especially debug along the whole sense - plan - act stack.

Also, today there exist also hybrid approaches. You take real recordings and abstract them into a simulatable format that you can then, e.g., use to variate and derive artificial scenarios for simulation. This can be used to analyse the influence of different situative paramters on the behavior of a function to pinpoint which parameter(s) caused certain troublesome behavior that have been observed in real drives.