Yes, because of no rear facing radar, I figure they need something more then monocular vision looking back, I haven't done the math to know for sure, but I imagine at highway speeds the gaps close pretty quickly if say your coming from an on ramp to the highway thats running full bore at 77 mph.
I think my biggest fear with the current arraignment is merge events and how they will be handled, it seems to me to be a huge blindspot in the system until the straightaway, as there is no radar and there no camera pointing like the b-pillars but towards the rear. But again, I'm not NN or vision guy, this is just my theories based on what I see, and more importantly what I don't see.
The more time goes on, the more I feel swindled, these cars including AP2.5 will never have FSD, it's just not going to happen. We "may" get lvl4 for highway driving, but I don't see this system taking us door to door, at least not without some serious hardware retrofits. I think the camera approach Tesla is trying to take can most certainly work, but for it to really work, they need more coverage of the rear of the vehicle, just not perpendicular to the backside of the car. And, sadly, I do think they need at least 1 rear facing radar, but that's probably for another thread.
Yes, I'm also aware of the depth from context stuff, it's actually the comma.ai job interview. You solve the problem, greater then .4 I believe and your in like Flynn. Thought, I've spent the last 30 minutes or so trying to find it, no luck.
But thank you for your insights, the deep expertise you bring to the community are very englightening!
Man, it's so nice to have a respectable exchange with someone who has a different opinion. Thanks for restoring my faith that people of good conscience can disagree amicably.
I wish there were a way to know what it's going to take to make FSD reality. The physics and computation are complicated enough but then we get all of this company posturing and PR and FUD making it impossible for regular people to know what to think. Even experts are bitterly divided about the essential requirements but mainstream articles are constantly presenting one view or the other as if it's the consensus in the field.
I think that the general sense that, if you had software that matched human capabilities you could get by with just the senses humans have isn't particularly controversial. But how hard that might be is plenty controversial. And then there are folks who argue that we shouldn't settle for 'human level' abilities because you might still get crashes but not a lot of people seem to seriously argue that we should not deploy good systems while waiting for perfect ones.
Of course we don't have human level software yet and we don't know when that's coming. Probably you can drive a car with a lot less than fully human capabilities but we don't know how much less. Adding hardware might make the job easier and there are many, many fans of adding hardware. I say 'might' because some problems definitely get easier in the lab when you add more sensors or more kinds of sensors, but increasing complexity is also a problem and so far it's not clear whether it gets more complicated faster than it gets more easier. About half the pros think the answer to that is a no brainer, but some of them think it's a no brainer that it's too complicated and some think it's a no brainer that more sensors always win. The best funded efforts throw lots of hardware at the problem because they can and they know that they can always cut back later. Of course, when you see all these 'leaders' out there with sensor festooned SUVs it gives the sense that a lot of hardware is needed, but really those guys are just trying out lots of things in parallel. It's still an open question what bits are needed and which bits are superfluous. And of course the 'leaders' want the problem to look hard and expensive to scare away competition.
But then maybe it really is hard and expensive. The jury's still out.
And there's this elephant in the room that nobody really talks about because it's kind of hard to explain, but getting along smoothly with other road users - drivers, bikers, and pedestrians - has turned out to be a much harder problem than anyone was expecting. As a human you can do a pretty good job of predicting the actions of other road users, but can software do that? Several years ago when google first started fielding cars in mountain view they discovered that not hitting anything wasn't good enough. Famously they got rear ended a lot because their vehicle didn't blend with traffic well and so they made this shift from trying to perceive the environment well enough to avoid obstacles to trying to interact smoothly and predictably. They've been focused mainly on that for years now and they still can't do it well enough to field a car in a real urban environment. Cruise tests in SF, which is probably the about the toughest environment that the U.S. has to offer but they have been very tight lipped about their real capabilities and the general sense is that it's not because they're doing unexpectedly *well*.
Eventually we'll have the 'kitty hawk' moment where someone puts something out there that more or less does the job and then the world will have one example of how to do it. And after that things will get better fast. But in 2006 when I was at the DARPA urban challenge I thought for sure we would see kitty hawk before 2016, maybe even before 2011. But now it's 2018 and I'm still waiting. Until we have that moment, and probably for a while after it, we won't really know. I'm still optimistic, but I don't trust my ability to predict it anymore.
The delays in EAP are demoralizing and lack of transparency from Tesla is not helping there. It certainly makes FSD feel really far away. I totally get that and I feel the same way. But I have this other window on the problem that not many people at TMC get to enjoy, and that's a detailed understanding of what's happening in AI right now. It's not an exaggeration to call the pace of improvement in AI techniques shocking and unprecedented. Internally the field is being turned inside out as all these 50 year old obstacles are finally being overcome. Newbies coming into the field are doing stuff that the old timers considered practically impossible just a couple of years ago. Every couple of months I see something happen that just takes my breath away. That 'breathtaking' pace of advancement makes me optimistic that we'll see solutions to FSD. And if the solutions are good enough then the hardware doesn't have to be overwhelming.
Going with cameras alone is definitely, *definitely* a bet that the software is going to get a lot better really soon. When I think about the magnitude of the bet that Musk is placing on this I've got to imagine that he's the bravest guy to ever run a large company - or the craziest. I could never make a bet like that.
It's a bet that might not turn out. But my feeling these days is that if the software doesn't get a lot better it's not going to matter how many sensors you have. Driving in the real world is a really hard problem.