Also Dojo isn’t some magical technology. It’s an FPGA platform which we have available today on AWS and hasn’t changed the game for deep learning use cases. It’s just edge optimised.
LiDAR however has changed the game. It works well and costs are rapidly coming down.
Want to wager? Have you seen the FSD beta videos? Looks like they've pretty much solved it for autonomous driving. Also my experience when using Tesla FSD with Navigate on Autopilot. It pretty much looks inevitable at this point.
Also Dojo isn’t some magical technology.
The magic, is how they won't have to label manually. It will automate the labelling.
LiDAR however has changed the game. It works well and costs are rapidly coming down.
Just like Scotty Kilmer when he says, "changed the game." Using cameras + AI will leapfrog it for applications like autonomous driving. What will be left, are applications where you can't afford too much processing. I agree with you that LIDAR will get cheap enough for that.
Also, you don't even address my main point about military applications and active sensors. I'm kinda proud of that bit of reasoning.
Have you? Plenty of videos where it fails pathetically (like https://youtu.be/antLneVlxcs). This looks nowhere close to “pretty much solved”. In fact, it’s an active danger to the rider and its surroundings.
The military one is interesting... seems like autonomous military vehicles will be massively prone to adversarial attacks and as soon as the line of training is discovered that entire line of hardware suddenly can’t see the mine it’s about to drive over, for example. I think unless you get to full general AI then you’re still going to need a human in the loop to protect against adversarial attacks for everything but the most mundane cargos, and at that level the difference between LIDAR and CV is probably minimal
https://news.crunchbase.com/news/lidar-startup-blackmore-rai...
Active sensors aren't viable anymore for military applications outside of severely asymmetrical conflicts. Active sensors will advertise their position to seekers and guided ordinance. Passive sensors are going to be required for military autonomy, and military R&D will guarantee these will supplant active sensors like LIDAR.
Let's say that Tesla goes 100% pacifist, and declares that no Tesla technology, like Dojo, will ever be used for military purposes. Just the fact that Tesla FSD had solved computer vision and the fact that Dojo exists will guarantee that someone will replicate those things for military applications.
Active sensors could still become so cheap, however, that they will be used in applications where each unit must be very cheap and therefore can't have the onboard processing for full-blown autonomy AI and computer vision. LIDAR could then be used to reduce unit costs for those super low unit-cost applications. These LIDAR would be much smaller and much lower power than the LIDAR for autonomous cars.