-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla autopilot HW3
- Thread starter verygreen
- Start date
Heck, I sleep in my tesla now as I recently told my wife she looks good for a mom who just gave birth , I’m proof that you can sleep in Fool Self Driving mode on the current firmware.
Actually nothing is easier than criticizing without offering a better alternative. I had no problems when my staff would complain as long as they had proactive input as to what needed to be done.
As far as myself, I know too little on a number of things to make absolute statements. My personal belief is that Musk is overly ambitious but believes what he says. I think he looks at humans and sees that they only rely on vision so he feels that should be enough. In theory he is correct. However, I feel (I'm no expert) that he way underestimates the human mind vs. current NN systems.
Some people have equated believing in Musk to believing in Trump. I am personally very insulted. I strongly believe I could do a better job than Trump. I do not believe I would have been as successful as Musk at either SpaceX or Tesla.
Short selling is like betting that a football team will lose. I have no problem if it is done quietly. What I do have a problem with is trying to influence the stock price. What many short sellers do is equivalent to betting on a team to lose and then intentionally exposing the players to the flu virus so they will be sick on game day.
diplomat33
Average guy who loves autonomous vehicles
Since this is the thread to discuss AP3, I thought I would share my short term expectations for AP3 and what I think Tesla will achieve on AP3 at least at first. Hopefully, Tesla will surpass these expectations in the long term. But these are my short term expectations (1-2 years from now):
Highway Driving
Nav on AP is already pretty good at a lot of highway driving and the recent removal of stalk confirmation is promising. And we know AP3 only uses 10% of its compute load now for Nav on AP which leaves plenty of room for it to do more. Hopefully, AP3 will increase the reliability where Tesla can safely start to remove some of the "hold the wheel" nags. So I think it is not unrealistic that AP3 can get us to L3 highway driving at least for most highway situations. I imagine Nav on AP could remove the "hold the wheel" nags for some situations and notify the driver in advance when they need to hold the wheel again when Nav on AP sees a scenario it might be less reliable in (like a complex interchange coming up based on GPS or seeing a lot of traffic up ahead).
Expectation: Highway self-driving with no nags in most situations
City Driving
With the increased computational power of AP3 and the larger, more complex NN that Tesla will eventually build, I don't think it is unrealistic to expect that Tesla will get good traffic light detection, stop sign detection and intersection detection on AP3. And EAP can already handle the mundane situations of staying in your lane really well. So I think AP3 will be able to automate a good deal of city driving (stopping at red lights, making simple turns etc). So the car will be able to drive itself in some city driving but with driver attention. I definitely think Tesla will keep the "hold the wheel" nags for city driving for awhile.
Expectation: City self-driving in some situations (with nags so driver pays attention)
Highway Driving
Nav on AP is already pretty good at a lot of highway driving and the recent removal of stalk confirmation is promising. And we know AP3 only uses 10% of its compute load now for Nav on AP which leaves plenty of room for it to do more. Hopefully, AP3 will increase the reliability where Tesla can safely start to remove some of the "hold the wheel" nags. So I think it is not unrealistic that AP3 can get us to L3 highway driving at least for most highway situations. I imagine Nav on AP could remove the "hold the wheel" nags for some situations and notify the driver in advance when they need to hold the wheel again when Nav on AP sees a scenario it might be less reliable in (like a complex interchange coming up based on GPS or seeing a lot of traffic up ahead).
Expectation: Highway self-driving with no nags in most situations
City Driving
With the increased computational power of AP3 and the larger, more complex NN that Tesla will eventually build, I don't think it is unrealistic to expect that Tesla will get good traffic light detection, stop sign detection and intersection detection on AP3. And EAP can already handle the mundane situations of staying in your lane really well. So I think AP3 will be able to automate a good deal of city driving (stopping at red lights, making simple turns etc). So the car will be able to drive itself in some city driving but with driver attention. I definitely think Tesla will keep the "hold the wheel" nags for city driving for awhile.
Expectation: City self-driving in some situations (with nags so driver pays attention)
So you are basically saying you expect to receive what Tesla already said they will deliver? You used a lot of words to say that...
diplomat33
Average guy who loves autonomous vehicles
The order page just says that FSD is navigate on autopilot, auto park, enhanced summon, traffic light and stop sign detection and automatic city driving which is not very specific in terms of what level of autonomy FSD will be. But past FSD descriptions have strongly implied full L4 autonomy. So I think it is worth using a lot of words to better describe exactly what my expectations are, and give reasons for my expectations, especially since I am not expecting FSD to be full L4 autonomy right away.
I know the conversation has moved past wonky visualizations, but I have a situation that beats even my previous image and it’s so ridiculous that I must share it. I’m again parked in the garage at work with a car to my left and nothing else around me. Behold, the view of the world through the eyes of my car: Stable visualizations secured - Create, Discover and Share Awesome GIFs on Gfycat
scottf200
Well-Known Member
I've looked at this a lot lately and my visualization (AP2.0 X) is really stable and accurate on real roads (i.e. not parked in a garage).
804am. My brain operates like that before 1 or 2 cups of coffee. How caffeinated was your car's NN?
Got an AP1 loaner today and I am pleased to report that it blows in comparison to my AP 2 car.... Jesus I’ve waited so long to say that
Honestly while driving the visualizations are pretty steady, with only the typical amount of oddities. It just amuses me greatly watching how crazy things get while parking.
Judging from the looks of things, my car opted for a more herbal start to the morning.
pilotSteve
Active Member
“For sale: never driven above 3.3v!! Like new!!So how much do you suppose they'll get for all that silicon on the scrap market?
lunitiks
Cool James & Black Teacher
I drove an AP 2.5 car a few days ago and I'm pleased to report that I was pleased to get back in my AP1 car
EDIT: Ok so it was cool to see all the visualizations (which were kind of choppy though), and I can't say I've tried NOA yet, but I maintain that AP1 feels far more reliable. I guess those episodes with phantom braking back in the good old "6 months maybe"-days still stick to my brain while driving AP2. So probably just in my head
lunitiks
Cool James & Black Teacher
I drove an AP 2.5 car a few days ago and I'm pleased to report that I was pleased to get back in my AP1 car
EDIT: Ok so it was cool to see all the visualizations (which were kind of choppy though), and I can't say I've tried NOA yet, but I maintain that AP1 feels far more reliable. I guess those episodes with phantom braking back in the good old "6 months maybe"-days still stick to my brain while driving AP2. So probably just in my head
I realized that my AP2 lane change red line warning and IC visualizations are becoming irreplaceable .... It's getting nuts.... AP1 doesn't do crests and valleys well, and the lane change on NoA is impressive but will take some time to figure out how to coexist and optimize.... Wishy there was an easier way to deny an automatic lane change without touching the screen.
Cirrus MS100D
Member
It really seems like getting their maps in order is the next big piece, especially for those of us outside CA. NoA PLEADED with me last night <crying> “For the love of GOD move over to the far left lane, we NEED to take this exit!!!!” It hammered me 10 times in a row and I pounded the ignore button (I should’ve just disabled it...) because the exit was right-lane only. This was on 95 / NJ Turnpike where it splits near the Lincoln Tunnel.
NoA can only make lane decisions based on the maps and if the maps are lousy, the car behaves like an idiot.
Are you sure this wasn't NoA wanting to go faster? My problem isn't a map issue. NoA will want to move over to the left to go faster when there is heavy traffic and the exit is a couple of miles away. Heck, at 1.1 miles it still wants to change lanes. I wish NoA could recognize heavy traffic and get over in the exit lane sooner rather than later when traffic will make lane changes difficult.
I agree about Maps but I don't think they will ever be good enough just receiving map data from the map data vendors out in the world. To truly make Map data relevant to FSD Tesla needs to be able to collect fleet Map data.
This would enable them to do things like mark locations of stop signs, lights and speed limits in certain areas based upon an approach which uses consensus across many cars driving in those paths.
This could also be used to fix bad maps over time by generating something like a heat map of where cars are driving to illuminate the navigable path.
Similar threads
- Article
