"You are right, that was a red light!"
At first I was skeptical, but reading the article, they do track object across video frames which for me it's the minimum a system like this should do. All previous FSD systems are just Fake self-driving.
"That proprietary AI driver identifies features in images and point clouds, groups them into objects, and tracks them across frames, time-stamped to the millisecond to account for differing frame rates. The AI thus builds confidence over time, acting on object detections that persist across several frames, rather than, say, slamming the brakes due to a camera blip on a single frame."
>they do track object across video frames which for me it's the minimum a system like this should do. All previous FSD systems are just Fake self-driving.
I'm pretty sure Waymo does this. It's hard to imagine not using this information, it's basically 50% of what I do when driving (e.g. that kid that was running on the sidewalk suddenly disappeared from view in an unexpected way; I wonder if he's about to cut across the street, etc).
I wander what it does for different objects.
Oups a ball crossed the street 15m ahead. Let's slowdown to a crawl and watch for the kid is what a human would do.
I suppose they may have special cased the ball case.
This is very basic for any self driving car system you can see on the road today or even 10 years ago.
> The AI thus builds confidence over time, acting on object detections that persist across several frames, rather than, say, slamming the brakes due to a camera blip on a single frame.
Nice in theory but in practice they still need a ton of training data. The bitter lesson is very bitter. Tesla's end to end FSD uses occupancy networks and brake stabbing still hasn't been fully solved.
> slamming the brakes due to a camera blip on a single frame
Ironically, I'm sure most humans have done the eyeball equivalent a handful of times, despite our knowledge of object permeance.