The Associated Press reported on August 28, 2024 that William Stein, a technology analyst at Truist Securities, has taken Elon Musk up on his invitation to try the latest versions of Tesla’s Full Self Driving system three times in the past four months.
During his ride, Stein said, the Tesla felt smooth and more human-like than with past FSD versions. But in a trip of less than 10 miles, he said the car made a left turn from a through lane while running a red light. “That was stunning,” Stein said.
He said he did not take control of the car because there was little traffic in the area, which made the maneuver seem less dangerous. Later the car drove down the middle of a parkway, straddling two lanes that carry traffic in the same direction. This time, Stein said, he intervened.
During two earlier test drives he took, in April and July, Stein said Tesla vehicles also surprised him with unsafe moves.
I use FSD daily (Boston). I am on v12.3. It is very good. My office commute is about 1 hour. I have 1 or 2 interventions every day.
This is down from 7-10 interventions from 2 months ago. I am waiting for 12.5x.
I have taken Waymo rides in SF. Waymo is also very good but does not go on highways and stays below a speed limit.
It would be safer (and less exhausting) for me to drive manually than to sit and wait for the 1 or 2 interventions in an hour long drive.
You really need to get this down to something like one non-lethal intervention per month before anyone is buying it.
I’ve always thought the geofenced interstate highway system would be the first application since it’s limited access and there’s less chance someone is going to roll a baby carriage in front of your vehicle. But I guess the risk analysts are afraid of hitting something at highway speeds.
The funny thing as we drive all we do manually is interventions. We steer this way and that to avoid people in parkinglots, people walking we see at the last minute, and anything else that we can kill or dent our car body with.
All of it is interventions.
Sentient life would be trust worthy.
Machines do not learn.
One billion miles later and the car is in the middle of two highway lanes. Not a care in the world because machines do not care.
I don’t think so. While all my car has is “lane follow” and “smart cruise”, I find that it takes away 90% of the aggravation of driving, at least on limited access roads. It’s easy to go for hours and hours and let the car do most of the work.
It’s just OK on more congested, less predictable streets, but even then it’s helpful, if not perfect. It, as the saying goes, takes a load off.
I find FSD is much safer and will not drive without it anymore.
The progress in last 7 years has been amazing. The pace is getting better and better. The progress in last few months has been incredible with v12.3 as game changer. It is magical. It maneuvers around construction zones, recognizes Police hand gestures, allows for dynamic route changes. I always thought it would be until 2030 that we will get true FSD. I am now highly confident that we will get it by 2026. I am looking forward to the v12.5x.
It’s not true. You have to trust me on that, at least until you try Tesla FSD for a while. It has become far more relaxing, AND the car is mostly a better driver for normal circumstances, it sees better than I do, it stabilizes in the lane better, and it is almost always predictable. About the only time I have to really be on alert is for “big” changes, like an exit from the highway where there are 4 lanes, some left, some right, and some straight (to a service road for example). I’ve had the car drive my wife and me to a local movie theater many times, that ride includes roads from 25 mi/hr to 55 mi/hr, it has right turns and left turns, some with arrows some without, it has 2 traffic circles, it has 4 “complex” intersections (2 left turn lanes, 2-3 straight lanes, 1 right turn lane, etc). And it succeeds every time with recent versions (last year or so). Once reaching the theater, I take over to park, but sometimes it can park on its own (shocked the he11 out of me the first time it did that!), but it doing the parking is too slow for me, and it doesn’t choose the parking spots that I prefer. It does not pull into my driveway by itself yet.
Also, you have to understand what “intervention” means in most cases. The VAST majority of the time it does NOT mean “I took over because it would have crashed” or “I took over because it was about to go through a red light”. Instead it means “I took over because the next lane over was moving much faster” or “I took over because I see it would be better to get into the turning lane early” or “I want to get around this darned truck quickly” or “I don’t like the parking spot it chooses and want to pick my own”, etc.
Here’s a report from a group of auto reviewers Automotive Marketing Cousulting, Inc. (AMCI).
LOS ANGELES, Sept. 24, 2024 /PRNewswire/ – After what is likely the most extensive real-world test of Tesla’s FSD ever conducted by an independent third party – covering more than 1000 miles (1610 km) – AMCI Testing’s result demonstrates just how far Tesla must go before Robotaxi operations can be safely undertaken. AMCI Testing’s evaluation was based exclusively on Tesla’s latest software iterations, 12.5.1 and 12.5.3, with time spent across four distinct driving environments: city streets, rural two-lane highways, mountain roads and freeways.
While impressive for a uniquely camera-based system, AMCI testing’s evaluation of Tesla FSD exposed how often human intervention was required for safe operation. In fact, our drivers had to intervene over 75 times during the evaluation; an average of once every 13 miles. "
We have no idea whether that’s representative or not, of course. But your anecdotal experience isn’t very optimistic for any near-term switch to full self-driving under FSD.
The average US driver goes about 660K miles between accidents, so an AV system probably has to get to about half a million miles between interventions to be usable as a Level 4 or 5 AV. You’re experiencing about four orders of magnitude more interventions than that. With FSD improving by less than an order of magnitude over the last year (for you and your drive, at least).
Of course, a key problem with a study like this is that they didn’t also evaluate any competitor’s system and other “studies” are so differently constructed as to make it impossible to really compare them.
If I am not mistaken, the comparison is to us, the average driver, not to another competitor.
Tesla can have the lowest rate of interventions and still be a long way off from level 4 or 5 autonomy.
Edit: Weird! The “system” is now auto-editing posts to remove quotes. I quoted tamhas in this post and for some reason the “system” took the quote out. First time experiencing that.
