Thread by @theckman

On December 15th I left to drive from Los Angeles to near Philadelphia. Got back yesterday, after a total of 6,392 miles in the Model S Plaid with #FSDBeta.

Drove more than 99% autonomously, and I've some not great @Tesla Autopilot + FSD thoughts and experiences to share... 🧵

Let me start off by saying I recorded most of the trip in 4k@60fps from a GoPro inside of the car, and so I _should_ have footage of many of the things I’m complaining about here. Might be hard to find it though, amongst all the footage.

The second is that even with the issues I’m raising, it did reduce the personal cost of driving so many miles. It was a lifesaver. That said, it caused stress in other ways that wasn’t present years ago, some of which I think are legitimate safety concerns for me and others.

I don’t know of another way to get feedback to the engineering teams at Tesla to address these issues, so you get this Twitter thread cosplaying as a novella.

My goal here is to have this system improve, because I do believe it has the potential for huge benefits.

The tl;dr is that I wish @Tesla spent time on achieving hands free highway driving instead of #FSD on city streets. As a result of changing focus, Autopilot experience is worse than when we got our Model 3 in summer 2019.

Also, FSD is exceptionally poor outside of California.

1. The removal of radar on the highway was a huge mistake. Tesla Vision very often misidentified vehicles in front as being much closer than they are, trigging strong phantom braking. Sometimes losing 20mph of speed before I can react, which is huge a safety concern.

To be clear this was when using Navigate on Autopilot, and was not the FSD stack. I was reminded of the reports of the Bay Bridge / Treasure Island Tunnel pile-up on Thanksgiving 2022.

nbcbayarea.com/news/local/inv…

1a. Even worse, sometimes it even thought it saw vehicles when nothing was present. I could see them rendering up on the instrument cluster, and then going away. This also was causing very strong phantom braking.

2. The Tesla Vision system is very poor at maintaining a safe follow distance in low visibility scenarios (fog + rain). Set AP to follow at 7 car lengths, and it was closer to 3 or 4.

I guess it was trained with mostly clear conditions. 👎

I’ve used Autopilot with Radar in similar conditions in the past, and it did a marvelous job of maintaining the correct follow distance. This feels like a huge regression, and I look forward to cars getting that new @Tesla HD Radar.

How about a retrofit, @elonmusk? 😉

2a. Somewhat related to Tesla Vision and low visibility. The automatic high beams also kind of suck. They’ve gotten better at turning themselves off in response to traffic, but they take 4-5 seconds to turn back on afterwards. It needs to be faster, especially on curvy roads...

This is especially painful in the Model S / X 2022 refresh, because it’s not super easy to turn on the high beams with the capacitive touch buttons on the yoke. It requires you to hold the capacitive touch button down for a few seconds.

3. Related to no radar, the car now really sucks at deciding whether or not it should move over to pass a car we’re approaching from behind. It will often change its mind, which will cancel the lane change even if you triggered it manually.

Radar is way more consistent.

4. Changing of maximum speed in response to a speed limit change is inconsistent. Sometimes it’ll change the speed to match, and other times it won’t make any changes. It also sometimes does it incorrectly, like going from 65 to 45 on a highway without a sign showing speed change

4a. Related, it also sometimes will reduce your maximum speed even though the speed limit increased. So if doing 70 in a 55, when the speed limit becomes 60 your max is then set to 65.

This makes no sense…

5. #FSDBeta outside of California was a fucking mess. Regularly choosing the wrong lane, trying to go straight through turn lanes, often misreading lane markings, struggling to make turns. Lots of disengagements.

I use FSD in LA all the time, and I was reluctant to use it in PA.

This isn’t even an exhaustive list, just what I could describe to fit within a single tweet. It also struggled to stop at intersections, often coming to a stop 50 - 100 feet before the actual intersection stop line.

6. The FSD stack (for roads, not yet highways) does not react quickly enough to the maximum speed being changed, including in observance of speed limit signs. It can take many many seconds to reach a legal speed, when going from a 55mph zone to a 35mph zone.

In fact, someone I know was pulled over on New Years Eve (yesterday) by a Nevada state trooper because #FSD took too long to slow down in observance of the speed limit going from 55mph to 35mph.

Is that acceptable? Especially when we’ll have single stack on the highway?

Unless they fix this, I think the V11 release with a Single Stack model is going to result in many people getting speeding tickets while the car is operating autonomously, because it fails to slow down fast enough.

At the end of the day, I think this stuff has tremendous potential. But at this point there needs to be focus and good execution, while not causing regressions in the experience especially on features that impact your safety and the safety of others on the road.