I recently watched a great talk by Andrej Karpathy about Teslas transition away from radar towards vehicle speed estimation using only vision: https://youtu.be/gZ2SeiLjaEc
I wonder if it would have prevented the tragic outcome here.
> In most of the video, the Tesla maintained a speed of 69 miles per hour, but just before impact it briefly increased to 70 m.p.h. then slowed in the final second, according to data from the car.
> Mr. Rajkumar of Carnegie Mellon, who reviewed the video and data at the request of The Times, said Autopilot might have failed to brake for the Explorer because the Tesla’s cameras were facing the sun or were confused by the truck ahead of the Explorer. The Tesla was also equipped with a radar sensor, but it appears not to have helped.
> “A radar would have detected the pickup truck, and it would have prevented the collision,” Mr. Rajkumar said in an email. “So the radar outputs were likely not being used.”
Radar throws a lot of false readings so there are many situations where it is ignored.
For examples, readings of stationary objects are mostly false.
My guess is that the speed difference was so big that also in this case Tesla decided that the radar reading was stationary and false.
I suspect its a vision or fusion problem--as the article points out, the lighting wasn't great.
Why doesn't the car use ultrasound for AEB? Perhaps risk of false positives coupled with the fact that its range is so short it would only have time for limited braking anyway?
See specifically at 2:20:22 [2]:
> We use that radar to annotate what the vision is seeing, the bounding boxes that come out of the neural networks, so instead of human annotators telling you that this car, in this bounding box, is roughly 25 meters away, you can annotate that data much better using sensors.
I take that to mean that the radar is used as part of supervised learning for updating the model offline, whereas just the cameras are used as input to the model.