Actually there are plenty of possible reasons, and the lack of economy of scale is only one of them.
As an example, clinical blood chemistry analyzers range in price from around $5,000 for a limited capability desktop unit to well over $200,000 for high throughput laboratory units. But surely with modern technology the cost could be reduced enough to make it affordable to consumers, right? Someone tried that. You may have heard of her - her name was
Elizabeth Holmes. Having an incentive to make technology more affordable is only part of the battle. You have to take physics into account too. Theranos failed not because Elizabeth Holmes set out to defraud anyone, but because there's a limit to what can be even done even with today's cutting edge technology.
Waymo has been using LIDAR for 10 years and has yet to achieve the economy of scale required to make it affordable. They build their sensors in-house because there isn't anything suitable on the open market. The result is very slow development of a niche-use technology. Cameras OTOH have seen huge improvements over the years due to their wide range of applications in consumer equipment, particularly cell phones. Due to the large market and demand for higher resolution they are far more capable today than 10 years ago. Other industries are leveraging that technolgy for their own purposes. Drones for example not only use cameras but also miniature acclerometers and GPS units that were developed for cell phones. That's why you can buy a toy drone which is more capable than machines that cost thousands a few years ago.
Musk made a bold play with camera-only sensing, and many said it wouldn't work. Now some are changing their tune as Tesla is proving that it
does work. The reason Musk is right and they were wrong is that he knew how to make it work. It's not the sensors, but the
software. Tesla could put more cameras and/or other sensors on their vehicles to get more ranging data, but it wouldn't help because the computer wouldn't be able to analyze it. At one time they had RADAR, but stopped using it because the combination of RADAR and camera data just confused the computer.
Now Waymo is talking about adding generative AI to their software. They have developed an End-to-end Multimodal Model called
EMMA to help with things like object detection and road graphs. However the image processing frame rate is low and
doesn't incorporate data from RADAR or LiDAR sensors. Tesla started doing that years ago, which is why they are so far ahead of Waymo now. Tesla's FSD can be used
anywhere, not just in a few geofenced areas which it has been explicitly trained for. FSD is a generalized solution that only needs camera data and therefore can do anything a human can, provided the 'brain' is smart enough.
Another factor affecting LIDAR which hasn't got much attention is the light pollution and interference that may occur when large numbers of vehicles are using it, including the potential for physical harm. Current LiDAR is running the maximum power it can without being dangerous, relying on the constantly changing angle to avoid frying eyeballs. However some people have found out to their cost that cell phone cameras can be destroyed by it. Laser beams don't diverge much with distance, so at several meters the spot is still small enough to burn sensitive things.
Once the general public starts to appreciate the danger of LiDAR you can bet there will be oppostion, and there
should be. Just because the beam normally moves too fast to fry your eyeball doesn't mean it can't happen. Imagine if every new vehicle had it. On a busy road there would be literally millions of invisible laser pulses shooting out in all directions. People already fear 5G cell phone signals which are much less powerful and have no known health effects, whereas you can prove the potential for harm from LiDAR simply by pointing a cell phone at it. They have the potential to damage the cameras in other cars too, and perhaps even other LiDAR units. There is also concern about the effect on wildlife, particularly birds. Lasers are currently being used to scare birds away from crops, but what about areas they are ligitimately inhabiting?
Bottom line is cameras are a safe non-polluting
passive technology, while LiDAR is inherently polluting and potentially damaging. Since roads are designed for humans, cameras should be enough. LiDAR is just a crutch to make up for a lack of image processing power.
