HW2.5 capabilities

[AnxietyRanger]

Hmm, is @calisnow and @buttershrimp a tag team?

Innuendo is easy to throw around. Good discussions about autonomous technologies much harder.

I think Bladerskb made some interesting comments above on what is needed from Level 4 is various scenarios. I don't know or care who he/she is, the opinions/views are posted and we can discuss them.

The idea that Tesla might implement FSD as a geo-fenced service in tightly mapped areas is an interesting one.

Sounds plausible. What is needed of the suite (or not) is where the diversity of opinions seem to begin.

Has the geo-fenced suburb idea been discussed before somewhere on TMC?

 

[AnxietyRanger]

@bonnie @HankLloydRight @Max* @ecarfan @EinSV @jeffro01 - if you have not followed this lately - @Bladerskb appears to have suddenly gone from months of sounding like an actual kid from Michigan - to someone who is not American - in a single post - post 932. Grab shots of it in case this entire discussion disappears. @AnxietyRanger has been running interference for blader for weeks - it almost looks like an assignment. AR is too smart to have made the decision to let someone else with an entirely different voice and geographic origin use the blader account - if it happened it wasn't AR's decision - to me this says they both have a handler - there is supervision and someone else is calling the shots. AR would not have been stupid enough to try to discredit @buttershrimp just now for first calling out what he noticed about blader's account - because he would know that simply points suspicion to him even more strongly. I'm convinced more than ever that AR is following orders from someone else. @Bladerskb is the technical mouthpiece and @AnxietyRanger is the persuader/guider/influencer who steers conversation, topics and uses subtle manipulation techniques to attempt to neutralize posters who disagree with the message - and especially who "mess" with @Bladerskb.

To be clear, this is pure fantasy from where I sit and I am surprised to see you posting such things. Obviously I am a mere Tesla car owner two times over with a long history on this forum - and I know nobody here in "real life". I have no idea who @Bladerskb is, but IMO he/she posts interesting and often technical views on autonomous cars. His/her views are valuable IMO independent of who is posting them. We need more people posting interesting technical views and data points, not less.

Frankly, the greatest irony of this exchange is that @Bladerskb above is actually posting a very valid reason - first time in a long time - to believe FSD and Tesla Network might actually happen on some form of AP2.

I wasn't sure I'd believe it (let's be honest, I have kind of given up n the FSD in my car), but @Bladerskb's message of geo-fenced, tightly mapped suburbs together with Waymo's recent video and taxi service beginnings certainly made it sound possible.

A Tesla Network to run the local kids to soccer practice within a neighborhood that Tesla has mapped to an inch? I can see that.

That is a valuable thought independent of whether or not we agree with Bladerskb on what exact technologies are needed to achieve it.

 

[calisnow]

To be clear, this is pure fantasy from where I sit and I am surprised to see you posting such things. Obviously I am a mere Tesla car owner two times over with a long history on this forum - and I know nobody here in "real life". I have no idea who @Bladerskb is, but IMO he/she posts interesting and often technical views on autonomous cars. His/her views are valuable IMO independent of who is posting them. We need more people posting interesting technical views and data points, not less.

Frankly, the greatest irony of this exchange is that @Bladerskb above is actually posting a very valid reason - first time in a long time - to believe FSD and Tesla Network might actually happen on some form of AP2.

I wasn't sure I'd believe it (let's be honest, I have kind of given up n the FSD in my car), but @Bladerskb's message of geo-fenced, tightly mapped suburbs together with Waymo's recent video and taxi service beginnings certainly made it sound possible.

A Tesla Network to run the local kids to soccer practice within a neighborhood that Tesla has mapped to an inch? I can see that.

That is a valuable thought independent of whether or not we agree with Bladerskb on what exact technologies are needed to achieve it.

What did you study friend? What drew you to linguistics? Or was it literature? Critical theory? Philosophy? Is it hard to make a living where you are? There is a kind of melancholy romanticism when conversing with a professional and you've cracked their armor just a tad. Just for a moment their eyes look at you, they flash - frightened - but also secretly delighted perhaps - that someone has recognized and in a way admires the expert level of their craft. It's human to appreciate recognition for a job well done. But those eyes can't flash for long, can they? They can't give much back to you except the smallest signal, the slightest nod, acknowledgment - that they appreciate your recognition - perhaps the click of a "funny" button. Then they must go back to work, back to character. "Must" - why? Too much at stake. Mouths to feed, people to answer to. Those f*ckers in the West - they'll never understand what it is to be us. It's easy for them to look down on us - what hypocrites Americans are - and delusional and clueless as well! It's easy to be sanctimonious when they have enough to eat, when they can support their families. But they don't even realize they're being hypocritical - those damn Americans are just - so - clueless! You are right, @AnxietyRanger - we are. Wherever you are - whoever you are - whatever your circumstances - I don't judge you for what you do. We all - make our choices. We all sell out to some degree. We are born - we think we'll change the world - then the world changes us. Somewhere I imagine you smiling for a moment, the edges of your lips curling up slightly, lines form for a moment above your eyes as your brow furrows a bit. Your eyes close, you let out your breath in a sigh. You open them again and for a moment they're not looking at your screen - they're down and to the right - a distant gaze - focused on an unseen horizon far way - your lips still locked in that half smile but the outer edges of your eyes sagging down. The moment's over - you must get back to work. The paradox of your work of course is that you must have an intuitive empathy to be good at what you do - a finely tuned humanity. And yet of course you must continually negate the self to do your work. I hope life works out well for you in all the ways that really matter. Goodnight.

 

[AnxietyRanger]

What did you study friend? What drew you to linguistics? Or was it literature? Critical theory? Philosophy? Is it hard to make a living where you are? There is a kind of melancholy romanticism when conversing with a professional and you've cracked their armor just a tad. Just for a moment their eyes look at you, they flash - frightened - but also secretly delighted perhaps - that someone has recognized and in a way admires the expert level of their craft. It's human to appreciate recognition for a job well done. But those eyes can't flash for long, can they? They can't give much back to you except the smallest signal, the slightest nod, acknowledgment - that they appreciate your recognition - perhaps the click of a "funny" button. Then they must go back to work, back to character. "Must" - why? Too much at stake. Mouths to feed, people to answer to. Those f*ckers in the West - they'll never understand what it is to be us. It's easy for them to look down on us - what hypocrites Americans are - and delusional and clueless as well! It's easy to be sanctimonious when they have enough to eat, when they can support their families. But they don't even realize they're being hypocritical - those damn Americans are just - so - clueless! You are right, @AnxietyRanger - we are. Wherever you are - whoever you are - whatever your circumstances - I don't judge you for what you do. We all - make our choices. We all sell out to some degree. We are born - we think we'll change the world - then the world changes us. Somewhere I imagine you smiling for a moment, the edges of your lips curling up slightly, lines form for a moment above your eyes as your brow furrows a bit. Your eyes close, you let out your breath in a sigh. You open them again and for a moment they're not looking at your screen - they're down and to the right - a distant gaze - focused on an unseen horizon far way - your lips still locked in that half smile but the outer edges of your eyes sagging down. The moment's over - you must get back to work. The paradox of your work of course is that you must have an intuitive empathy to be good at what you do - a finely tuned humanity. And yet of course you must continually negate the self to do your work. I hope life works out well for you in all the ways that really matter. Goodnight.

...oookay. One more time: My education, professional life is nowhere connected to cars or anything in cars, nor do I have any interests or ownership in anything related to cars or car companies, nor have I ever had - beyond owning many cars, of course (and yes, many of them German premiums, but not only). I assume I am much less biased in that way than many on TMC, because many do own automotive stocks - I do not, never have. (If I did, I would probably go long TSLA, because I believe Model 3 will succeed.) Nor is my education or professional life connected to anything related to cars, communications, lobbyism, activism, literature or anything of the sort that you imply there. I am here as a person interested in cars and technology as a hobby - on my free time, without a dime.

In an interesting twist, given that the Russian connection was suggested by @buttershrimp and I guess by @calisnow above, I was just actually reading a review of Hillary's book (on my reading list) and there was this quote: "When you know why you’re doing something and you know there’s nothing more to it and certainly nothing sinister, it’s easy to assume that others will see it the same way. That was a mistake."

This quote and conversation helps put some of the Internet in perspective. When we don't know each other, it is easy to assume and even build up caricatyres and characters of others, that may not in reality have anything to do with who they are - yet still, somehow, our confirmation biases end up reinforcing our assumptions time and time again. It is perfectly understandable, though unfortunate of course. The saddest part of it is, there is nothing much you can do to change those mistaken perceptions once they have formed.

It was actually @buttershrimp who, unintendedly I guess, made this point when he/she said he/she would apologize if his/her comments about other member's motivations turned out to be wrong. That's the beauty of such a promise on an anonymous forum. You will never have to apologize for such an accusation, because you can never truly know whether or not you were wrong. I can say you are wrong, which @buttershrimp you are when it comes to me (of @Bladerskb I have no idea), but given that you doubt my word, how could I ever change your mind.

I have lived most of my adult life with the thought that never attribute to malice, what can be explained by incompetence - or coincidence. I am sure it makes me naive at times, but mostly I think helps. I assume @Bladerskb is genuine and any flaws are merely human, but I don't know for sure - probably never will. But most importantly, I don't think anything anyone posts on TMC at this stage can have any material effect on Tesla or TSLA, so I don't think it matters either way. If someone really trolls here, what a monumental waste of time IMO.

Some interesting thoughts about HW 2.5's ability were presented by @Bladerskb, I do think it would be to the benefit of this thread to discuss them - no matter who we think he/she may be.

Just to set the record straight and be a record.

If HW 2.5 has corner radars they will be able to achieve L4 (real L4 not basic) highway autonomy in 2021.
Their computing system will be sufficient without even tapping into the reserve. Remember they have about 4 TLFOPS and Mobileye's eyeq4 has 2.5 Tflops and ME will use two of those for highway autonomy. So Tesla should able to achieve L4 self driving with it.


HW 2.0 however is dead in the waters.

Also if HW2.5 has corner radar they will be able to achieve L4 self driving in suburb neighborhood and communities with the additional gpu (8 TFLOPS). Suburb neighborhood have pristine road marking and wide lanes and no pedestrians, cyclists, constructions and road debris.

Therefore Tesla will start using its Tesla network around 2022 in suburb areas they have mapped and heavily validated.
Then they will allow owners to use their car for ride share.
Sorta like what voyage is doing.

Voyage’s first self-driving car deployment – Voyage

Urban traffic however will require lidar and musk will come to his sense in 2020 and this was what i meant by all L3+ cars will require lidar. I'm speaking of urban environment.

 

[lunitiks]

My condolences @verygreen

 

[calisnow]

A brief sampling of the persuasion/debate/influence techniques of @AnxietyRanger:

What gives you away @AnxietyRanger is the contrast between your obvious intelligence and skill with language - and the absurdity of some the positions you are forced to defend because of the position you are locked into. Also with observation your techniques begin to crystallize because of the sheer volume and rate of your posts:

• Professional level restraint - almost superhuman - WRT your lack of defensiveness when accused of something. You know that being defensive is perceived as a sign of guilt - so you do a perfect job of balancing between maintaining your position and not getting flustered/personal about it.
• Diversion as a tactic to focus attention where you need it to be and away from where you do not. You are intelligent enough to avoid directly arguing points that are futile - your genius that you pick your battles while still continually pushing the same position:
• Example 1 of diversion: @Bladerskb is now on the record as appearing to be a team account - someone clearly American wrote using that account for months. Then all of sudden a stunning slip happened and that handle appeared in the span of one post to change to an Eastern European voice. You stepped in to defend him - but you did not attack the obvious - that that is not the same person using the handle as before - because you know it's a losing argument. So - instead - you simply counterattack by questioning the character of @buttershrimp - accusing him of using innuendo - again - this is a diversionary tactic because you are not dumb enough to argue points you can't win. The weakness of diversion is that it is not a normal human response, psychologically, to ignore shocking/stunning surprises. If a person *does* ignore something that is rather shocking (but which would be counterproductive to one's message - which requires a perception of authenticity to be persuasive) - instead of having conversation around said shocking event - one simply steps around that little doo-doo and hopes to focus attention on something else.
• Trojan horse "agreements" with your interlocutor: Expert level use of using minor, extraneous agreement with your interlocutor as a trojan horse to cloak a larger point - which is always on message with the idea that Tesla is behind.
• Example 1 of trojan horse: You try to neutralize @buttershrimp's discovery of the bladerskb handle's "team" appearance *not* by attacking or trying to tear down my analysis of the language of blader's post (because you are again, too intelligent to fight what you can't win - seriously, if you were a general you'd be on your way to five stars) but instead by trying to claim that bladerskb suddenly agrees with the idea that Tesla will get to FSD with its camera system - neglecting to mention of course that the key point of the "Russian" bladerskb's post was to push forward a claim that Tesla's FSD technology is year and years away (like 5 years). 2021 or 2022 is course far away enough that it would create a valuation problem - or at least doubts.
• Obviously absurd claims with no support: This part is you being careless.
• Example one of absurd claim you defend: The idea that TMC is unlikely to have an effect on stock price when in fact TMC is where much technical discussion takes place - in public - in a world in which TSLA is the largest short on the NYSE - with something like 27% of Tesla's stock currently shorted. Leading shorts are all over the media publicly trying to change opinion - the idea that with billions on the line, professional influencers would not be at work is - naive. Any short (or group) with the capital would of course hire writers/influencers - the salary would be small potatoes vs what is at stake. But you have to defend the idea that TMC discussions are irrelevant to stock price.

Diversion

 

[AnxietyRanger]

What gives you away @AnxietyRanger is the contrast between your obvious intelligence and skill with language - and the absurdity of some the positions you are forced to defend because of the position you are locked into.

No, that is just the difference of our opinions.

Any restraint I show is a an attempt to respect the thread. It really was a good one. I hope it returns.

 

[Matias]

Everybody’s discussing computer vision and sensor configuration (cameras, radars, lidars, bla bla bla) as if it was the holy grail and fundamental challenge to solve L3, L4, L5 or “FSD”. It’s like people think that the more sensors, the closer you are to driverless.

Let’s not forget that sensing is only 1/3 of the AV equation. Let’s not forget that there are actually three fundamental tasks to solve here:

1. Sensing, aka perception;
2. Localization, aka reasoning;
3. Driving policy, aka planning/forethought and negotiation.

We’re really just scratching the surface with these "radar vs lidar" discussions. We’re farting around like Statler and Waldorf with stuff that’s really not interesting anymore. Perception is pretty much solved now (the issues being fine-tuning, cost-lowering and distribution; "technicalities" IMO).

View attachment 252507

Anyway, with sensing, we all know you can do collision avoidance, lane holding and traffic aware cruise control. Your car sees all the things – lanes, pedestrians, street lights, signs (“free space”, “objects”, “driveable paths”) – so you can cruise pretty safely. The car won't kill you. Great.

But that’s not self-driving, is it? That’s not all it takes if you want to “read a book” while your car takes you from your garage to work – no controls touched?

You can equip your car with Bill Gates’ worth of lasers, 32K cameras and yottaFLOP GPUs, but your car won’t have a clue where it is! Without localization – i.e. super high precision maps – you'll have to tell your car which turns to make (intervene), which lanes to go to (intervene), which exits to take (intervene), where to park (intervene), etc. Localization – pin-pointing your location; having your car understand and reason about where it is on the road, and in the world – is absolutely critical to achieve true self-driving.

I think most people get this, but I think people often forget that this is a monumental task.

First off, you need resolution. GPS won't cut it, but with lidar, radar and/or great structure-from-motion algorithms you can get pretty decent res. maps. So, send some cars a few times down the road with decent sensors and you'll get an HD map. Great.

But how useful is a nanometer precision map once the terrain changes? I mean, you got these super fancy sensors and this super fancy map that you made, and then wham! Construction! Wham! Temporary detour. Wham! You can’t go that way anymore!

View attachment 252504

So you see the grand mapping challenge is actually updates. I mean really, ultra-high refresh rates. A three month old HD map is no good. A month old HD map isn’t reliable either. I think you want at least daily updates to achieve reliability. You have to be able to trust the maps.

The problem is, such extreme refresh rates require an enormous fleet of vehicles doing that mapping work. I know what you’re thinking: Tesla is the OEM with the biggest fleet of HD-mapping-capable vehicles. They’ve got 260 Model 3’s for Gods sakes! (Ok, that was under the belt I own a Tesla so I can joke about it.)

But the question is – is Tesla’s fleet big enough for the mapping necessary for FSD?

Well Mobileye certainly doesn’t think so. Mobileye believes the real solution is crowd sourced, cross-OEM, truly massive data harvesting from pretty much every car out there. Mobileye thinks you need to leverage ALL the cameras and internet connections that’s standard in every new vehicle built today. So, not a few hundred thousand cars. Millions.

Quantity is obviously key to the mapping challenge. Will Mobileye succeed before Tesla? Well I tend to think so. They’ve signed 27 OEMs, most notably VW, BMW, Ford, General Motors, Nissan, Volvo, Audi and Hyundai. That. Is. Effing. Huge.

We shall see, shan’t we @AnxietyRanger?

Now, finally we have driving policy. Planning. Car interacting with humans. Signaling to other drivers and pedestrians what you’re intending to do next. Receiving and interpreting such signals from other agents. “Giving and taking”, negotiating. This is kind of the event horizon of autonomous driving. Nobody’s got this figured out AFAIK. Probably takes ridiculous amounts of machine learning to get there, and my guess is as good as yours.

But let's stop dwelling on radars and lidars. Let's talk maps and planning. It's going to be fun, I promise.

Now watch this again:

Humans can drive without precision maps? If I'm thrown in a foreign country, I can still drive by using plain old paper map?

 

[AnxietyRanger]

Humans can drive without precision maps? If I'm thrown in a foreign country, I can still drive by using plain old paper map?

I would assume precision maps are a bit like radar and Lidar. Sure, they can be used to add extra senses to self-driving cars that are not there in a human (seeing through objects for radar, seeing in pitch black dark for Lidar, "seeing" around the corner for precision maps), but mostly they all are about offsetting weaknesses of computer vision...

Precision maps (as well as radar and Lidar) add a layer of redundancy to what the computer is "seeing". They are its sanity check, if you will. Precision map making also include possibilities of precision training driving policy, I might imagine...

Hence I actually think the notion that FSD in geo-fenced neighborhoods might be the first non-motorway Level 4/5 activity on the market. It is something I might actually see Tesla being able to do with something akin to current suite(s). If they know intimately where they are driving, making sense of it - even with limited sensors - becomes much more easier.

 

[bonnie]

I've been asked to weigh in on the kerfuffle erupting. Forgive me for some rambling, on my first cup of coffee.

I had an issue with AR discussing me (instead of posts) in the past. In fact, that tendency to pick me apart is probably at the root of why I have minimal interaction today. However, that said ... I'm going to do that. I have reason to believe that AR is who he says he is. I think his Pavlov tendency to defend an outlier has gotten him drawn into this.

If I had to hazard a guess (and certainly his private info to share), I'd guess he's likely a scientist of some type (or should have been). My experience in managing scientists vs. engineers points to something like this. Engineers were straightforward. Scientists wanted more data, more data MORE before being willing to commit to an opinion. So I learned to manage that. (Anyone managing scientists, pm me. I have advice. All works great!)

@AnxietyRanger, not that you're asking for my advice -- but if, in your musings, you were willing to attach a 'level of confidence' to the different things you say, it might make a difference. Instead of saying 'we must consider all sides' and implying that all sides have equal weight, how about saying 'I have a low level of confidence that this might be the answer, but am unwilling to throw out the possibility'? Or 'I have an approximately 80% level of confidence, having [this] data would allow me to raise/lower my confidence'. I think that approach would be better tolerated by the general forum & would allow you to keep all opinions on the table - but properly weighted so that people actually can see what your opinion is.

As for the other person in this discussion ... I've been convinced for a long time that he has other (clear) motivations for showing up on this forum. I have zero idea regarding the theory being thrown out here, but would ask this: What difference does it make if he's a disgruntled former Mobileeye employee or part of a managed team? The result is the same. And by responding to him, no matter who he is, you give him validation.

(On the other hand, I recognize the 'fresh meat' aspect of all this -- hard to argue that one post is written by a different person. So don't let me ruin the fun!)

 

[BigD0g]

Can you folks bitch at each other elsewhere? I'll even create the thread for you, STAY ON TOPIC!

Thread De-railers bitch about each other here.

 

[lunitiks]

Does anyone have anything remotely on-topic to contribute in here??? Where are the mods for christ sakes

 

[AnxietyRanger]

@bonnie I appreciate you posting and will try to take your input to heart!

@lunitiks I did post several thought about the geofence above, any comments?

 

[BigD0g]

Do we start the HW3 thread now, it appears Xavier has landed.

Nvidia unveils new supercomputer for level 5 autonomous driving

 

[mrkisskiss]

maps maps maps maps.

It's all about maps. I'm pretty convinced at this point that's basically the next step in EAP's development. Whether or not this is waiting on Mapbox Drive to mature, or they're starting with that and adding their own layer on top, or just doing their own thing - it's required to make any kind of real advancement.

The argument that humans can drive without a map is a bit weird. For example, I may be able to drive and not crash without a map... but this is different to driving safely and driving well. If anyone from out of town tries to drive across London, I guarantee they'll mess it up. They'll survive, sure (well, hopefully), but they'll do all sorts of stuff wrong - get stuck in box junctions trying to turn right, drive in bus lanes, be constantly in the wrong turn lane, screw up complex merges, get stuck in weird one way loops, stop on red routes etc etc. But, once they've done it a few times, it'll go pretty well - because they now have foresight and prior knowledge. I would definitely want my car to have this.

HD maps can be updated in near realtime in many ways, including harvesting from the fleet. It's a useful form of redundancy. If the cameras fail, I'd be willing to hedge my bets that the road looks the same as it did a few minutes ago. With hazard lights flashing, and the horn blaring, it could probably pull over relatively safely.

Bottom line: I can see how you get to autonomous driving with surround cameras and a map. Just cameras + radar? No way.

Btw, NVidia just announced their DrivePX Pegasus, coming H2 2018. It has 10x the power, 320 TOPS, with a lot less draw than the Drive PX2. NVidia reckon that'll give the power they need for Level 5 robotaxis. I think it's clear that Tesla won't be able to achieve FSD with their current hardware - regardless of it being HW2 or HW2.5. But, they can probably give a pretty cool Highly Automated Driving experience, which is good enough for me.

 

[Pale_Rider]

Nvidia is always going to say their latest and greatest (in this case not even available for a year...) is what you HAVE to have to do what you want to do. You will never find them saying that their last product is sufficient and there is no reason for them to continue developing something new to sell you... According to them, 2 PX2 chips could do FSD. Now duel Xavier's are required for FSD. They have a product to sell and it will always be the only way to do what you want if you listen to them... That said, I'd be more than happy for Tesla to upgrade my AP2 computer with this new Pegasus board.=)

 

[AnxietyRanger]

I think if Tesla does upgrade AP2 or AP2.5 cars one day (if needed and if it happens at all), it will with their own silicon development - once that is done one day year(s) down the road...

 

[Matias]

maps maps maps maps.



The argument that humans can drive without a map is a bit weird. For example, I may be able to drive and not crash without a map... but this is different to driving safely and driving well. If anyone from out of town tries to drive across London, I guarantee they'll mess it up. They'll survive, sure (well, hopefully), but they'll do all sorts of stuff wrong - get stuck in box junctions trying to turn right, drive in bus lanes, be constantly in the wrong turn lane, screw up complex merges, get stuck in weird one way loops, stop on red routes etc etc. But, once they've done it a few times, it'll go pretty well - because they now have foresight and prior knowledge. I would definitely want my car to have this.

But avoiding that does not need any HD maps or positioning. Normal GPS is enough. Why do you need centimeter accurate maps and navigation, that is my question.

 

[Cirrus MS100D]

maps maps maps maps.

It's all about maps. I'm pretty convinced at this point that's basically the next step in EAP's development. Whether or not this is waiting on Mapbox Drive to mature, or they're starting with that and adding their own layer on top, or just doing their own thing - it's required to make any kind of real advancement.

The argument that humans can drive without a map is a bit weird. For example, I may be able to drive and not crash without a map... but this is different to driving safely and driving well. If anyone from out of town tries to drive across London, I guarantee they'll mess it up. They'll survive, sure (well, hopefully), but they'll do all sorts of stuff wrong - get stuck in box junctions trying to turn right, drive in bus lanes, be constantly in the wrong turn lane, screw up complex merges, get stuck in weird one way loops, stop on red routes etc etc. But, once they've done it a few times, it'll go pretty well - because they now have foresight and prior knowledge. I would definitely want my car to have this.

HD maps can be updated in near realtime in many ways, including harvesting from the fleet. It's a useful form of redundancy. If the cameras fail, I'd be willing to hedge my bets that the road looks the same as it did a few minutes ago. With hazard lights flashing, and the horn blaring, it could probably pull over relatively safely.

Bottom line: I can see how you get to autonomous driving with surround cameras and a map. Just cameras + radar? No way.

Btw, NVidia just announced their DrivePX Pegasus, coming H2 2018. It has 10x the power, 320 TOPS, with a lot less draw than the Drive PX2. NVidia reckon that'll give the power they need for Level 5 robotaxis. I think it's clear that Tesla won't be able to achieve FSD with their current hardware - regardless of it being HW2 or HW2.5. But, they can probably give a pretty cool Highly Automated Driving experience, which is good enough for me.

I totally agree with this. Conceivably, one should be able to "dead-reckon" a car (with an HD map) from point A to point B, so long as it can:
1) Avoid obstacles (e.g. other cars, pedestrians)
2) React to changes in the environment (e.g. traffic lights)

Obviously there is still tremendous work to be done with 1 and 2, but getting 1 and 2 to 90% and HD maps to 90% should offer a very high degree success.

 

[Matias]

I would assume precision maps are a bit like radar and Lidar. Sure, they can be used to add extra senses to self-driving cars that are not there in a human (seeing through objects for radar, seeing in pitch black dark for Lidar, "seeing" around the corner for precision maps), but mostly they all are about offsetting weaknesses of computer vision...

Precision maps (as well as radar and Lidar) add a layer of redundancy to what the computer is "seeing". They are its sanity check, if you will. Precision map making also include possibilities of precision training driving policy, I might imagine...

I believe this is right. You need HD maps and HD positioning, IF your machine vision is not on par with human. If you have as good computer vision processing as human, normal precision GPS and maps are enough.