571–580 of 582 posts
Earlier quoted context omitted.
How does Waymo handle it now? There are many videos of Waymo depots with dozens of cars not running into each other.
Lidar critics like to pretend that anti-collision is not a well-studied branch of Computer Science and telecoms. Wifi, Ethernet and cellphones all work well simultaneously, despite participants all sharing the same physical medium.
Earlier quoted context omitted.
Just for the record though, Musk isn't blindly anti-LIDAR. He has said (and I think this is an objective fact) that all existing roads and driving are based on vision (which is what all humans do). So that should technically be sufficient. SpaceX uses LIDAR for their docking systems. I would argue that yes, we do use vision but we get that "lidar depth" from our stereo vision. And that used to be why I thought camera…
Sufficient if all else were equal. But the human brain and artificial neural networks are clearly not equal. This is setting aside the whole question of whether we hope to equal human performance or exceed it.
In fact, that's why radio/music/podcasts thrive. Because we're bored when we drive. We have conversations, etc. We daydream.
As long as the skills relevant to actually driving are on parity with humans, the rest doesn't matter.
In fact, in a recent podcast, Musk mused that you actually may have a limit of how smart you want a vehicle model to be, because what if IT starts to get bored? What will it do? I found that to be an interesting (and amusing) thought exercise.
Earlier quoted context omitted.
> Tesla are producing cyber cabs now which are 10th the price of Waymo's and can drive autonomously anywhere in the world. Wait what? when did they actually enter mass production? > I mean humans have Lidar sensors Real time slam is actually pretty good, the hard part is reliable object detection using just vision. Tesla's forward facing cameras are effectively monocular, which means that its much much harder to get…
> Tesla's forward facing cameras are effectively monocular Notably, human perception is effectively monocular in driving situations at distances of 60 feet or farther. It's best in the area where your limbs can reach. We don't need stereoscopic vision to drive.
crucially we have excellent implied depth, and object detection, something that even non-realtime state of the art tracking doesn't have.
human depth is much more complex than just parallax, which some poeple use as an argument that "pure vision" monocular depth is possible to do robustly. It will be, but not for a while. Especially as depth is only part of the problem. object categorisation is the other.
Earlier quoted context omitted.
> comma.ai proves that lidar is red herring I mean it doesn't. If you actually look at it comma.ai proves that level two doesn't require lidar. Thats not the same as full speed safe autonomy. whilst it is possible to drive vision only (assuming the right array of cameras (ie not the way tesla have done it) lidar gives you a low latency source of depth that can correct vision mistakes. Its also much less energy intens…
It's fair to point out that comma.ai is SAE level two system, however it's not geofenced at all, which is an SAE level 5 requirement. But really that brings up the fact that SAE's levels aren't the right ones, merely the ones they chose to define since they're the standards body. A better set of levels are the seven I go into more detail about on my blog. As far as distinguishing shadows on the road, that's what rada…
The geofence is there to keep a system inside its safe operating area. The point of level 5 is that it doesn't need a pre-computed map to be safe, it can safely and autonomously navigate an unknown area.
Earlier quoted context omitted.
> Tesla are producing cyber cabs now which are 10th the price of Waymo's and can drive autonomously anywhere in the world. Wait what? when did they actually enter mass production? > I mean humans have Lidar sensors Real time slam is actually pretty good, the hard part is reliable object detection using just vision. Tesla's forward facing cameras are effectively monocular, which means that its much much harder to get…
Not mass production yet, but the first one rolled off the completed assembly line at giga texas last week Sensor fusion is not far simpler, when the sensors disagree, and they will often, you have to pick which to trust. It is amazing to see how many people here are confident they know the one true way to build autonomous systems based on nothing but wanting to confirm their biases
Sensor fusion is a piece of piss. How do you think any of the VR headsets work? SLAM is compute expensive, and runs at ~30hz, sensor fusion is the only way to give a smooth experience.
Online calibration is actually the hard part, even then its largely solved.
Earlier quoted context omitted.
Then you deeply underestimate how difficult the problem is, and deeply misunderstand where all the effort has been spent in developing autonomous vehicles.
If all the effort has been spent in trying to replicate the human brain then I am comfortable saying that is a mistake. We have a tool that can tell with great accuracy how far away an object is. The suggestion that we should ignore it and rely on cameras that have to guess it because “that’s how humans work” is absurd, frankly.
The problem of understanding what is happening in front of you while driving is preposterously more complicated than just a point cloud of distances. That is .01% of the problem. To solve the remaining 99.99%, you need interpretation of photons and sound waves into a semantic understanding that gives you predictive power to guess how the physical world will evolve and avoid breaking the rules of the road. Show me a mechanized way of understanding the causes of how the physical structure of the world is about to evolve, and I'll show you something that is imitating a human brain, however poorly. The cameras give you _plenty_ of data to determine 3D structure, at a higher resolution than the laser, without being emissive, for cheaper. It's a completely reasonable approach to focus your limited computational hardware on interpreting the data you have instead of adding more modalities with their own limitations that (according to nature) are demonstrably unnecessary.
The world is more complicated than slogans and pitchforks and Elon Bad.
Earlier quoted context omitted.
This is not quite true. It depends what you're talking about. Automotive LiDAR sensor prices typically range from $150–300 today for standard units. Mid-range ADAS systems (Ls+/L3) sensors are about $600-750 and the long-range units used by robo-taxis like Waymos are about $1,500–6,000 or more per sensor https://www.fleetowner.com/technology/article/55316670 The ~$75k per sensor in 2015 refers to the long-range senso…
I think the sensible cost to take is the latest cost in a Waymo model, as that's what "works" for automated driving [1]. I see numbers anywhere between $10K-$30K for that. [1] to a limited extent in a limited area - it's inaccurate to say that Waymo vehicles are fully autonomous.
So that's $2-7.5k per unit.
Earlier quoted context omitted.
Why make things more complicated than they need to be? Humans don't have lidar and we are the only intelligence that can reliably drive. Lidar just seems like feature engineering, which has proven to be a dead end in most other AI applications (bitter lesson). https://www.cs.utexas.edu/~eunsol/courses/data/bitter_lesson...
This knee-jerk reply is old and tired, and the counterarguments are well-trod at this point. Even if cameras-only can build a car that’s as good as humans, why should we settle for “as good as“ humans, who cause 40,000 fatalities a year in the US? If we can do better than humans with more advanced sensors, we are practically morally obligated to do that.
Phrased a different way, I would expect lidar to help marginally, but it is safer driving in general that will bring down fatalities. This could be done with cameras.
Earlier quoted context omitted.
Then you deeply underestimate how difficult the problem is, and deeply misunderstand where all the effort has been spent in developing autonomous vehicles.
If all the effort has been spent in trying to replicate the human brain then I am comfortable saying that is a mistake. We have a tool that can tell with great accuracy how far away an object is. The suggestion that we should ignore it and rely on cameras that have to guess it because “that’s how humans work” is absurd, frankly.
They get into accidents because they make bad decisions and get distracted.
If AI makes better decisions and don't get distracted, the amount of accidents will already be greatly reduced compared to humans.
Having lidar in addition to cameras will be of marginal benefit (but a benefit to be sure) when you realize what is actually important: proper modeling of the environment. And for this, cameras are better at providing than lidar, so you still will want cameras anyways.
The focus on lidar is really a red herring. You merely push the computational budget you have to understanding a point cloud instead of vision. You're back to square 1 of "how can I properly model the environment given this sensory modality". This is the part that essentially needs human level understanding of the world that you're missing.
As the other commenter says, you deeply misunderstand the problem.
Earlier quoted context omitted.
Just for the record though, Musk isn't blindly anti-LIDAR. He has said (and I think this is an objective fact) that all existing roads and driving are based on vision (which is what all humans do). So that should technically be sufficient. SpaceX uses LIDAR for their docking systems. I would argue that yes, we do use vision but we get that "lidar depth" from our stereo vision. And that used to be why I thought camera…
Humans aren't using only fixed vision for driving. This is such a tiresome thing to see repeated in every discussion about self driving. You're listening to the road and car sounds around you. You're feeling vibration on the road. You're feeling feedback on the steering wheel. You're using a combination of monocular and binocular depth perception - plus, your eyes are not a fixed focal length "cameras". You're moving…