The question is do the economics of self-driving cars work when you have to add and integrate all these additional equipment. Correct me, if I am wrong but aren't LIDAR cars supposed to cost you $200K+ ?
Tesla is imitating humans in a way that removing LIDAR and relying on the compute to make up for them and build more accurate picture.
Also, isn't Cruise owned by GM - who are the VCs here?
Lidar costs have dropped massively and continues to drop. Waymo, for example, claimed 4 years ago they were able to reduce cost of their Lidar by 90% from $75,000 to around $7500 [1]. In the meantime, range and resolution of these sensors have increased. Anyone not making use of Lidar in this day is just hamstringing themselves.
200k isn't that much for certain use cases, like shared cars. NYC taxi medallions were significantly more expensive yet the revenue from taxi rides was high enough that people kept buying medallions.
Not exactly apples to apples. A $200k hardware device depreciates in value over time, whereas the taxi medallion was a fairly liquid, appreciating asset.
If you search for "bosch lidar" there's a bunch of 2 year old news about them selling one designed for autonomous vehicles for $10,000, with statements that they could likely drive the price towards $200 with mass production.
> Tesla is imitating humans in a way that removing LIDAR
They are trying to imitate a fraction of what humans can do. And the state of the art ML research still does not account for issues like whether a photo of a person on a truck is real or not.
You really need LiDAR for accurate bounding box detection.
I recall a few years ago, GM/Cruise acquired a LIDAR manufacturer in 2017. Not sure if it has worked out, but their rationale for the acquisition for vertical integration makes sense.
Strobe’s solution will reduce the cost of making these sensors by 99 percent, Vogt says. “The idea that lidar is too costly or exotic to use in a commercial product is now a thing of the past.”
Humans don’t regress per-pixel depth or use convolutions and region proposals to draw bounding boxes around objects. They don’t function on models with fixed weights trained by backprop either. The idea that “vision only” somehow more closely resembles how humans drive quickly falls apart if you inspect the internals of these systems.
The similarity is in the problem to be solved, not the details of the compute pipeline. Depth must be inferred somehow, rather than measured by actively interacting with the surface, as it is in LiDAR.
If you really wanted to make this argument, you wouldn’t even want to bother inferring depth, since that’s not what humans do, not directly at least. If you’re actually trying to obtain a depth map as part of your pipeline, LIDAR (or LiDAR + vision feeding into a denser depth prediction model) would always be a better strategy, cost aside, since determining depth from images is an ill posed problem.
My claim is that humans use their eyes as a primary input for driving. I don't think it's controversial. We don't let eyeless people drive. Eyes do not shoot out lasers.
I think the comparison comes from the fact that humans infer a 3D map from stereo vision whereas LiDAR to some extent gives you the ground truth 3D map. You’re right that it falls apart pretty quickly though.
Except Tesla isn’t inferring a 3D map from stereo vision either, at least not outside of the front facing cameras - they’re using monocular depth prediction.
Neither are humans. Our eyes are so close together there's almost no disparity between the eye images beyond a handful of meters. We do 3D by inference beyond the near-field.
Tesla is imitating humans in a way that removing LIDAR and relying on the compute to make up for them and build more accurate picture.
Also, isn't Cruise owned by GM - who are the VCs here?