hoeschjp
Member
AP has more detail on the dimensions of vehicles than the visualization shows (and tracks a lot more objects than is shown as well).That’s a really good summary. And I wonder if they ever get there with the current set of sensors and cameras. When I look at the lane visualization I can see it cannot properly detect the size of a truck so it just plots a truck 3D model that moves weirdly as it passes you on the side. I get it; it has to guesstimate depth frame by frame based on 2D source files. But it has to make decisions on just these sensors. How far can you go with software if the hardware is limited?
I think what is needed is an additional Lidar sensor that can complement the existing data.
They currently only have one 3D model for a tractor, be it bobcat or pulling a short or long trailer, and they do not dynamically scale these. The center of the model has to be put somewhere within the bounding box, so it will always be goofy / off until they can represent these in a more advanced way (separate tractor with a scaleable or removable trailer or something). None of this affects the computer’s driving decisions though. It needs to improve IMO, but is just there for occupant awareness / confidence in the end.