-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD rewrite will go out on Oct 20 to limited beta
- Thread starter diplomat33
- Start date
powertoold
Active Member
What specific issue with the FSD beta do you think would take 3 years to improve?
DanCar
Active Member
Nothing specific, just avoiding accidents except for every 150K miles. Wife had an accident running into a black curb on a black street at night. I've seen stop signs that are mostly occluded. What happens when teenagers prank tesla with adversarial attacks? Put some tiny white tape on a stop sign? Many interesting scenarios. What happens when a toddler runs into a street from a park chasing a ball? Most people would be on guard. What if there are bushes mostly occluding the view of toddler entering the roadway?
powertoold
Active Member
We’re doing just fine then. The likelihood of the incidents you’re referring to is much lower than we’d like to imagine, and I’m sure the car can respond to humans running into streets. Karpathy showed several instances of the car’s emergency braking for pedestrians that were impressive given the reaction speed and prior occlusion of the human.
DanCar
Active Member
My estimate / guestimate is three years. What is your estimate / guestimate for no accidents on average every 150K miles?
Last edited:
Yes, people will trust it more, that is why it needs to achieve an acceptable "driverless accident rate". In other words, you have to assume the driver will not pay attention, would the accident rate still acceptable?
We don't know what magic number is. Tesla just says "safer than humans". So presumably, when Tesla's FSD achieves Tesla's internal metric, Tesla would probably remove driver supervision.
We know Waymo is willing to deploy their cars in driverless mode in a geofenced area with an accident rate of 1 crash every 338k miles and 1 incident every 130k miles. And we know that the crashes have been minor, no serious injuries, and most accidents were the fault of the other driver, not Waymo. I think it really depends on what types of accidents, Tesla's FSD gets into. If Tesla's FSD causes accidents or gets into accidents that regulator think the FSD should have been able to avoid, that would be a problem. And if the accidents are minor, that would probable be more acceptable. Serious injuries and fatalities would be less acceptable.
It really depends on what accident rate Tesla is "ok" with. But I think it also depends on what accident rate, regulators like the NTSB and the NHTSA are "ok" with. If they get mad, they can shut down FSD.
Tesla could minimize the risk by geofencing FSD to areas where it has high confidence. That way, FSD would only operate in areas where the safety rating would be the highest. And it would also mean less miles and therefore statistically less accidents. If Tesla deployed FSD in the entire US with no geofencing to hundreds of thousands of cars, considering all the edge cases that maybe FSD can't handle yet, it would most likely result in a lot of accidents. I am assuming Tesla would probably want to avoid that situation.
If you start off with 1 intervention every 7 miles and every two weeks the number reduces by 1/3, than after 6 updates you are at one intervention every 156,865 miles...
powertoold
Active Member
It's crazy I even think this, but 6-9 months lol. We've seen essentially every human maneuver from this beta, even unintuitive or difficult to program ones like:
Overtaking a slowly moving leaf blower on a two lane street
Moving over slightly to give room for a person opening a car door
Stopping for pedestrians intending to cross the street
Anticipating possible trajectories of running pedestrians
Moving slightly to the right because oncoming car slightly crosses line (at night)
Able to recognize an oncoming bicyclist at night, with just one visible flashlight
DanCar
Active Member
That math is bit off.
Tesla 1/3 improvement every month in interventions
My assumptions are: start at 1 intervention every 5 miles. Every month improve by 1/3rd. 37 months before reach 150K miles intervention.
Last edited:
BobbyKings
S85D > M3P > S Plaid
If you would assume an average 16 mile average intervention rate now, and a doubling of that every quarter then within 4 years you would achieve that.
Lots of assumptions though, but assuming exponential improvement seams ok.
DanCar
Active Member
I'm willing to take a bet on that.
Tesla will not release level 3 or 4 wide release for cities in that time frame or that published city intervention rate is more than 150K miles by that time frame.
I don't think the rate of improvement is linear though. Right now, improvements are coming fast because FSD is still early. As FSD gets better, solving those rare edge cases, gets harder and harder. Getting a 10% improvement gets harder and harder.
I feel like when I watch these fsd videos it's at the earliest stage of development not the final stage a lot of people are taking this as. It's all over the place and doing nothing consistently all the time. All of these issues being discovered by normal untrained people should have been cleared out prior to release. These aren't even difficult scenarios.
I think this was pushed out too early as a marketing gimmick. Looking at what the cars are doing I yea they are working on it we knew that but they haven't run into the model 3 driving into a parked car scenario or curbing its own wheels or all these bad left turns. 6-9 months no way. 3 years maybe but it's not going to be totally sleep in my back seat driving.
Also why haven't they improved the other features? They just stopped working on summon or reverse summon? I feel like reverse summon would be an incredibly useful feature.
I think this was pushed out too early as a marketing gimmick. Looking at what the cars are doing I yea they are working on it we knew that but they haven't run into the model 3 driving into a parked car scenario or curbing its own wheels or all these bad left turns. 6-9 months no way. 3 years maybe but it's not going to be totally sleep in my back seat driving.
Also why haven't they improved the other features? They just stopped working on summon or reverse summon? I feel like reverse summon would be an incredibly useful feature.
I imagine it will look like this.
The rate of improvement starts off fast but then slows down as you get closer to 100%.
I love how if you are critical of anything tesla in this forum it's well hated lol.
I'm sorry fsd is cool but I can't grasp how it's being taken as more advanced than waymo or cruise. They are literally full self driving already.
I'm sorry fsd is cool but I can't grasp how it's being taken as more advanced than waymo or cruise. They are literally full self driving already.
Yes, I checked it. My math was off. I did it through normal calculator by pushing six times “=“ sign. My bad. Using a normal spreadsheet I come to 35 updates.
But still. The FSD beta is out two weeks. They now achieve 1/3 less interventions after two weeks. This with only a handful of users.
1/3 improvement is not sustainable with only this handful of users, but by gradually having more users online it could be. 35 updates equals 70 weeks which is about 1.5 years.
Mo City
Active Member
Initially, I thought it was purely a marketing gimmick too. While I still believe marketing is part of the equation, the improvements in .11 and .12 convinced me there is more to it.
Summon (particularly reverse summon) remains risky in the extreme because there is no driver behind the wheel to take control. For that reason, I expect it to continue trailing behind the curve of overall FSD improvement.
Yea you are right. Nobody in the car for summon. Maybe they stopped working on it for now because of that.
DanCar
Active Member
What good is something you can't use versus something you can? When do you think you'll be able to use a Waymo or Cruise? When do you think you'll be able to use a Tesla in FSD mode?
Something that I can use is more advanced than something I can't use.
Last edited:
Similar threads
- Question
