TLDR:
What driving conditions have you found challenging for FSDb?
- FSDb is the best driver assist package available today, but is absolutely not ready to take unsupervised control of your vehicle.
- It would work better (and possibly more safely) with a more advanced driver monitoring system and making it hands free. High def maps might improve things to allow this degree of automation also.
- It struggles with turn lanes. This is probably better in areas with better mapping, but if its market niche is its ability to work on all types of roads, they need to do a better job with this.
- It allows closer proximity to other vehicles than I would as a human driver. I also don’t trust it in high traffic/merge situations yet, but this may partly just be me - it hasn’t really made many errors in this situation yet.
I’ve been skeptical of FSDb, and still am. That said, it has served the exact purpose for which I bought it.
I use my Tesla for commuting. It’s about 65mi each way, a mix of limited access highway (currently with heavy construction), rural open access divided highway, and city streets. This use case is why I invested in FSD. It is the only driver assist package currently available (other than possible comma 3) that can handle all driving tasks in all of these conditions. Competitor’s “hands free” driver assist features require pre-mapped limited access highways, thus none of them would work for the vast majority of my commute and would be effectively useless. Likewise, adaptive cruise control with lane keep (as I have in my other vehicle) is a far cry from the functionality of FSDb, requiring constant driver input. FSDb is technically a Level 2 assist feature, but it is so much more. And this is my biggest complaint with it.
Tesla needs to do a better job with the driver monitoring system for FSDb, so long as it’s a level 2 system. The difference between FSDb and adaptive cruise control, as far as my own experience, is that FSDb effectively locks you out of driver controls, unless a significant forced override is applied to the steering wheel. This makes the torque steering wheel sensor a pain in the a**. I’ve gotten better at applying the right amount of torque to the wheel to avoid the alerts, but in a system where my steering inputs are not translated into navigational change, I spend most of time monitoring myself to stop the nags, as opposed to actually paying full attention to what the system is doing. Personally, I think a better solution would be an advanced driver monitoring system to assure you’re paying attention, and make it hands free, lowering the steering torque required to turn the system off (ie when you place your hands on the wheel). In fact, I’ve had a number of instances where the force required to knock FSDb out resulted in a significant deviation in trajectory, usually occurring when it gets closer to a truck or other vehicle than I, as a human driver, am comfortable being, or when it misjudges speed or sharpness of a turn on city streets - both less than ideal situations to have a sudden jerking steering input.
Generally, I am very happy with the system. In fact, I see it as the best safety feature I could have. My job is shift work. I frequently drive home in the middle of the night, or the morning after a night shift. I risk falling asleep while driving all the time. It is a risk of the job. FSDb is by far the best driver assist feature for a tired driver. Though I haven’t fallen asleep at the wheel (yet), I have tested the system by squinting my eyes (so it thinks I’m asleep), and it pretty rapidly starts beeping at me. I also know that it can competently manage 95% of all driving tasks, so if I do fall sleep, it will keep on doing its thing while also monitoring me to make sure I don’t fall sleep. Definitely the best setup available for such a situation I have found. My only fear, is that if I nod off, when it beeps at me, I make a sudden steering input that results in the system being turned off and me loosing control of the car. Overall, I think the benefits far outweigh this risk.
I’m not interested in discussion of the ethics of my driving case and safety circumstances, that’s a personal decision. I am, however, fascinated by the performance of FSDb in my challenging use case. Overall, I think it does a fantastic job in difficult situations. Low light. Rural roads with absolutely no high-def mapping. The biggest errors it makes are mistaking turn lanes for new lanes of travel. There are a couple places on my drive where it does this almost every time, threatening to drive me off the road into the grassy median at 65mph. I’ve gotten more confident in the system to let it get closer to making a mistake, in fact letting it get in one of the turn lanes where I knew it extended past the intersection in a sort of merge lane. It correctly slowed, then merged back over once through the intersection, not that it should have been in that lane in the first place. At another tricky spot where the turn lane ends at the top of a rise (so it can’t see what’s ahead), it tries to take the turn lane almost every time, so I’ve learned to get out of the left lane at this location to avoid the situation. It also has the obnoxious habit of traveling in the left lane of the rural, non-limited access highway. This is not how I would drive as a human driver and I’m not sure why it is programmed for this to be the default.
Overall, FSDb is pretty good at taking all driving tasks away from the driver. It does struggle in certain situations and is absolutely not ready to take unsupervised control of your vehicle, unless you’re OK with it driving you off the road at highway speeds. A better driver monitoring system in combination with high-def maps and making it hands free would be a better solution, and get it closer to full autonomy. Here’s to hoping? I’m super excited to see what v11 and single stack brings. I like the city streets version better than the NoA version; it drives more like a human driver and has much faster responses, but also seems to be more prone to rash decisions (like getting into turn lanes inappropriately). We’ll see how it does on limited access highways…
I am curious what driving situations others have found FSDb to struggle with, or if people don’t have these issues in other areas with better mapping.
Image source: Full Self-Driving Computer Installations | Tesla Support Other Europe
Image added for Blog Feed thumbnail
What driving conditions have you found challenging for FSDb?
- FSDb is the best driver assist package available today, but is absolutely not ready to take unsupervised control of your vehicle.
- It would work better (and possibly more safely) with a more advanced driver monitoring system and making it hands free. High def maps might improve things to allow this degree of automation also.
- It struggles with turn lanes. This is probably better in areas with better mapping, but if its market niche is its ability to work on all types of roads, they need to do a better job with this.
- It allows closer proximity to other vehicles than I would as a human driver. I also don’t trust it in high traffic/merge situations yet, but this may partly just be me - it hasn’t really made many errors in this situation yet.
I’ve been skeptical of FSDb, and still am. That said, it has served the exact purpose for which I bought it.
I use my Tesla for commuting. It’s about 65mi each way, a mix of limited access highway (currently with heavy construction), rural open access divided highway, and city streets. This use case is why I invested in FSD. It is the only driver assist package currently available (other than possible comma 3) that can handle all driving tasks in all of these conditions. Competitor’s “hands free” driver assist features require pre-mapped limited access highways, thus none of them would work for the vast majority of my commute and would be effectively useless. Likewise, adaptive cruise control with lane keep (as I have in my other vehicle) is a far cry from the functionality of FSDb, requiring constant driver input. FSDb is technically a Level 2 assist feature, but it is so much more. And this is my biggest complaint with it.
Tesla needs to do a better job with the driver monitoring system for FSDb, so long as it’s a level 2 system. The difference between FSDb and adaptive cruise control, as far as my own experience, is that FSDb effectively locks you out of driver controls, unless a significant forced override is applied to the steering wheel. This makes the torque steering wheel sensor a pain in the a**. I’ve gotten better at applying the right amount of torque to the wheel to avoid the alerts, but in a system where my steering inputs are not translated into navigational change, I spend most of time monitoring myself to stop the nags, as opposed to actually paying full attention to what the system is doing. Personally, I think a better solution would be an advanced driver monitoring system to assure you’re paying attention, and make it hands free, lowering the steering torque required to turn the system off (ie when you place your hands on the wheel). In fact, I’ve had a number of instances where the force required to knock FSDb out resulted in a significant deviation in trajectory, usually occurring when it gets closer to a truck or other vehicle than I, as a human driver, am comfortable being, or when it misjudges speed or sharpness of a turn on city streets - both less than ideal situations to have a sudden jerking steering input.
Generally, I am very happy with the system. In fact, I see it as the best safety feature I could have. My job is shift work. I frequently drive home in the middle of the night, or the morning after a night shift. I risk falling asleep while driving all the time. It is a risk of the job. FSDb is by far the best driver assist feature for a tired driver. Though I haven’t fallen asleep at the wheel (yet), I have tested the system by squinting my eyes (so it thinks I’m asleep), and it pretty rapidly starts beeping at me. I also know that it can competently manage 95% of all driving tasks, so if I do fall sleep, it will keep on doing its thing while also monitoring me to make sure I don’t fall sleep. Definitely the best setup available for such a situation I have found. My only fear, is that if I nod off, when it beeps at me, I make a sudden steering input that results in the system being turned off and me loosing control of the car. Overall, I think the benefits far outweigh this risk.
I’m not interested in discussion of the ethics of my driving case and safety circumstances, that’s a personal decision. I am, however, fascinated by the performance of FSDb in my challenging use case. Overall, I think it does a fantastic job in difficult situations. Low light. Rural roads with absolutely no high-def mapping. The biggest errors it makes are mistaking turn lanes for new lanes of travel. There are a couple places on my drive where it does this almost every time, threatening to drive me off the road into the grassy median at 65mph. I’ve gotten more confident in the system to let it get closer to making a mistake, in fact letting it get in one of the turn lanes where I knew it extended past the intersection in a sort of merge lane. It correctly slowed, then merged back over once through the intersection, not that it should have been in that lane in the first place. At another tricky spot where the turn lane ends at the top of a rise (so it can’t see what’s ahead), it tries to take the turn lane almost every time, so I’ve learned to get out of the left lane at this location to avoid the situation. It also has the obnoxious habit of traveling in the left lane of the rural, non-limited access highway. This is not how I would drive as a human driver and I’m not sure why it is programmed for this to be the default.
Overall, FSDb is pretty good at taking all driving tasks away from the driver. It does struggle in certain situations and is absolutely not ready to take unsupervised control of your vehicle, unless you’re OK with it driving you off the road at highway speeds. A better driver monitoring system in combination with high-def maps and making it hands free would be a better solution, and get it closer to full autonomy. Here’s to hoping? I’m super excited to see what v11 and single stack brings. I like the city streets version better than the NoA version; it drives more like a human driver and has much faster responses, but also seems to be more prone to rash decisions (like getting into turn lanes inappropriately). We’ll see how it does on limited access highways…
I am curious what driving situations others have found FSDb to struggle with, or if people don’t have these issues in other areas with better mapping.
Image source: Full Self-Driving Computer Installations | Tesla Support Other Europe
Image added for Blog Feed thumbnail
my pants. Even when it does break the law, I usually let it just do its thing and I’m too lazy to correct it. I see a tremendous improvement from the stupid test score which I cheated with the reset hack compared to today.
