2019.40.50.1 Visualization observations

2019.40.50.1 Visualization observations

1) Stop signs seem to be picked up universally. In my testing, it picked up signs even when they were partially obscured by parked cars or vegetation -- included one I might have missed myself. It did have something of a challenge with a stop sign that was raised up on the end of an open gate. It seemed to feel that was a sign in the distance. On the very interesting front: It did not display a sign warning of a stop sign ahead as a stop sign, even though it was very similar.

2) All lights are stop lights, even those that are oriented horizontally or have just a single flasher, seem to get picked up and displayed as a vertical stoplight. This isn't at all a bad thing. It shows that it can collect all the variations and treat them as a class of objects. It seems to accurately display the color of the light no matter the orientation.

3) Trash containers ... I can only assume that Tesla has been using trash containers as obstacles in training. Otherwise, I'm not sure why this semi-random object would be visualized. Or why the system would be so insanely good at picking them up, even when they're 20' back from the road, or overturned.

4) Road signage. At first I thought the display was just imaging what's on the road, but then I went over a combo marker for right turn / straight and the display broke it into two symbols. Which makes me think that it's doing far more than just imaging. It's interpreting these symbols. It also seemed to make at least one mistake. For one clearly marked straight ahead arrow, it displayed a strange "X" symbol with a loop over it.

5) Parking lot arrows. The system also seems to pick these up, though it did have some trouble with arrows that were badly worn. QUESTION: Has anyone with Smart Summon seen the reading of arrows play into how it's responding in this release?

jamilworm | December 26, 2019

Very interesting observations! Thanks for sharing.

notify_craig | December 26, 2019

The fact that it recognized horizontal traffic lights is very encouraging. I (had) suspected that the reason we in Canada don't seem to be blessed with these latest visualization is that Tesla hadn't yet accommodated some Canadian provinces' traffic lights, which are horizontally mounted. Perhaps that isn't the issue but regulations. In any event, OP, thanks for sharing.

stingray.don | December 26, 2019

Recognizing trash bins is important for navigation on side streets. Trash bins can sometimes wind up in the roadway, and an autonomous vehicle needs to be able to recognize and avoid them.

rxlawdude | December 26, 2019

Does the trash can recognition register Taycans? :-)

Devilstower | December 30, 2019

Actually I'm assuming that the system is actually seeing an interpreting a lot of objects. Things such as roadside mailboxes and roadside hedgerows are certainly being picked up.

What's interesting about the trash container visualization is that it's being shown to the driver. Which does make me think they were used in some way to check the behavior of the system.

stingray.don | December 30, 2019

I think the significance of recognizing trash bins is that it is not uncommon for them to wide up being an obstacle in the roadway. For surface streets, FSD will need to recognize and avoid trash bins just as it does traffic cones.

M-A-B-MCMLXXX | December 30, 2019

Where are you guys seeing these? Is FSD a prerequisite?

dsscott0267 | December 30, 2019

I've noticed a few weird things in my area. Train crossing lights show as regular stop lights, anything that is small red or orange shows as a yellow cone (ex. slender post with orange top marking gas line, fire hydrant painted red, but ones painted other colors don't display). Also, sometimes the traffic light flashes red & green even though it is red. A set of traffic lights at a fire station that normally flash yellow just show as blank traffic lights.

Still pretty good visualization in my opinion.

dsscott0267 | December 30, 2019

Forgot to mention that I have a SR+ no FSD.

gene | December 30, 2019

I wonder - I have a MR with FSD, but hardware 2.5. Do you suppose the system is still doing everything it's doing in HW3, but just not showing the visualization? Is it still recognizing all these objects and reporting back for training?

Switchmon | December 30, 2019

SR+, FSD, HW 3.0

Seems like many of the written words and arrows are being interpreted accurately. Traffic lights colors are captured almost in real time.

One enhancement that I noticed from the previous release is the traffic cones. Traffic cones and traffic sticks (the white lane reflector sticks) are distinguished for the most part.

Some utility boxes are confused for trash cans if they are too close to the curb.

Scrannel | December 30, 2019

This morning mine thought a pumpkin on the side of our driveway was a traffic cone. Can it see orange?

M3phan | December 30, 2019

@ gene, I believe so, yes, in a shadow mode of operation. I think I recall that this was referenced in the autonomy day keynote.

EVRider | December 31, 2019

You need HW3. If dsscott is correct, FSD is not required.