@menachem 5780 Zoox was founded in 2014 by Australian designer Tim Kentley-Klay, and Jesse Levinson, who was developing self-driving technology at Stanford: en.wikipedia.org/wiki/Zoox_(company). The Tesla lawsuit alleged theft of line drawings of a warehouse layout, and had nothing to do with self driving, software, or Autopilot: www.theverge.com/2020/4/15/21222763/tesla-zoox-startup-lawsuit-settlement-autonomous-driving
Seeing all the complexity that ZOOX can navigate through is very impressive, and shows the importance of using lidar, radar, camera and mapping, and having the 360 degree view.
Roger Zimmerman: Yes. It's just ludicrous how Musk just denies any of the multi-sensing modes matter, AND detailed HD mapping (which Mobileye has a brilliant solution for, re crowdsourced updating with a narrow bandwidth data stream to keep the maps fresh), and the Tesla fanbois just blindly lap it up, even though Tesla is WAY behind in complex situations. What's really going on re real world performance, often illustrated by UA-cam videos, tells the real tale.
Hi everyone. Im really interested why the stationary car have grey smaller boxes inside, and moving don't? These grey boxes come probably from lidar - are they just filtered out for moving object for rendering? Or there is a reason for that?
Samuel Obirieze: Watch videos on Mobileye or read about Mobileye. They have a crowdsourced solution, which provides realtime updates. It's so good they have a business where they sell data to various cities, and plan to expand that massively. It's brilliant, and makes the usual Tesla claim that HD maps are a problem a thing of the past.
Karmesh Madhavi: Per the video, the car understands gestures. It stopped for a pedestrian when the pedestrian held his hand up. No reason it couldn't stop for police. And given how poorly many people drive and don't pay attention, I would trust that system by the time it's ready for real robo-taxis with no safety driver far more than I would the average driver.
Alper ALT: Re the driving, they ARE doing it live. And it keeps getting better. Re demo videos, what does it matter if they enhance the video later to make it easier for people to understand? What matters is that the CAR can understand things and handle them properly in real time. This is light years ahead of Tesla, given what UA-cam shows Tesla doing in complex city traffic situations. Compared to situations like what Zoox and Mobileye are routinely handling, Tesla is FAR away from even level 4 (at a reliable taxi level) for complex city driving, regardless of what Musk may claim. Waymo One is harder to gauge. They can do level 4 for a small and simple scenario by avoiding the more complex routes and behaviors (like difficult left turns). So unless Waymo One has lots going on they aren't showing, I think they're well behind.
@@rogergeyer9851 I was not aware they were broadcasting these driving events, live. Tesla is comparable to Openpilot I believe. Both using cameras and radar, only. Zoox is doing everything right. And I don't know about mobileye's progress.
On 21:49 there are grey blocks in the middle of the road, that probably comes from traffic lights (over the road) - and they again are rendered on the ground level). These appear just for one cycle - interesting, if they have problems with underdrivable objects.
This machine learning technology is amazing. Still not sure about how cost efficient that lidar & everything is. Tough to mass produce. seems very precise. Mapping streets seems life a flaw in teaching the AI to respond to unique scenarios. Roads are always changing
Mayko Marquez: But in the real world, there are solutions. Like the Mobileye crowdsourced, real time updates to its HD maps, which also gives it lots of data about city status (accidents, water leaks, fires, etc) which it is selling to cities for revenue -- plus the maps help the car drive much better. Musk just ignores all that, claims vision will solve everything, expects everyone to believe him, and is DEAD WRONG, re what happens vs. his predictions, every single year, re autonomous driving. I think pretending like Tesla doesn't have VERY serious competition and in fact that Tesla isn't seriously behind for complex driving, is just delusional. But of course, Musk won't admit this as it would likely significantly hurt Tesla's stock price. But over time, the real world will demonstrate what's up with working robo-taxis at level 4 from the likes of Mobileye, Zoox, and possibly Waymo, and Tesla will always be "working on it", unless Musk and team wakes up to the reality.
WizNitro 07: Plus, the premapping has a lot of serious advantages. Like better understanding all the traffic lights in complex situations and which apply to which roads. And, Mobileye has a very cool solution for the need for HD maps, updating the maps dynamically through crowdsourced cars, with very low bandwidth data streams. So it's really not much of a disadvantage to do the mapping if done right, AND all the advantages make the driving much more accurate. In a video like this, where the streets are crowded and narrow, clearly accuracy is ABSOLUTELY essential to safe driving over time. Seeing videos like this and like what Mobileye is doing, and comparing them to the chaos we see from Tesla in complex city situaitons (where FREQUENT driver intervention is required, and a frightening number of times, the car would cause a crash or even bad crash without that intervention, tells me that Tesla is WAY behind on this vs. the leaders, regardless of what Musk (and the Tesla fanbois) are claiming. And Tesla might get to the level 5 ungeofenced solution some day, but it sure as he** won't be 2021, and I don't see it happening before about 2030 on a truly global level. Musk should smarten up and be honest and look at a city-at-a-time solution for robo-taxis. Of course, to do that, he'd have to ADMIT that Tesla will have LOTS of city by city competition, and there goes the assumptions about unending massive Tesla income from high robo-taxi revenue, which will hit TSLA. So there's that.
They're carefully pre-mapped (though as explained at 18:30 they rely on cameras to detect and reason about the current state of the lights). Traffic lights don't change often, figuring out the association between light and lane in real time is nontrivial to do 100%, and you *really* don't want to hit edge cases like twitter.com/derekteay/status/1244450876774907904.
@@TaylorA: And if the car can do things like that reliably (helped by the HD mapping, no doubt), it's a better driver than MANY drivers. I've missed such signs driving in a strange situation the first time for example. And of course, see MANY, MANY people ignoring or missing such signs -- even the simpler ones that just say something like "no turn on red".
At 18:39 the second object behind ego car is extremely unstable - I understand, that it is occluded for almost all sensors, but can be that this instability is caused by interplay of camera detection and ML based tracking? Pretty interesting effect.
Great stuff; I'm also interested how this would function at 10 PM at night, where camera (i.e., visual) data is less reliable and radar and LiDAR is more important.
would be interesting to see how Zoox managed flight risk in a tight street such as around lombard street. How does it handle flight risk from very young children or small pets such as dogs? this is currently a headache for humans, requiring 100% alertness and immediate response.
Henry Lim: If you watched and listened to the whole video, by implication, they made it quite clear. Planning, wide field of view, LOTS of sensors, and being damn careful. Pretty much the opposite of Tesla in its current state, BTW.
Funny, that when some part of pedestrian is occluded, the boxes on abstract visualization becomes shorter, but still are placed on the ground level - I know, it is just visualization, but rises questions about quality in other areas. ;-)
Please start Zoox taxi service in Scotland. I can't stand not having the app to hail my Zoox ride. I want to stop giving my money to mafia run taxi services and mafia run bus drivers. Come on Zoox we know you can do it. kind regards LM p.s. Would a city full of these vehicles emitting radar harm the surrounding environment? Example, would a flower sat in a pot at the side of the road on a shelf survive its usual season, or would the fine leaves weather due to exposure to these radar emissions? Flowers never lie, they either survive or don't.
Imagine a world where autonomous cars had always existed and now people were arguing that humans should now also be allowed to drive. I think if you showed them this video, people would call you insane, to imagine that humans could navigate this.
I dunno about that. Cruise has a video out that looks exactly like this. Waymo has had this for years. Tesla is just ramping up their solution, which doesn't require maps. What baffles me is what's taking these mapped companies so long to mass produce and start servicing if it's as good as they're showing here.
Three Modalities - CAMERAS, RADARS AND LIDARS - STAGGERING 18 Cameras, 10 Radars AND 8 LIDARS - "This many sensors will be pretty expensive and UGLY if you had to pay for them on your own car" They are competitors for Waymo/uber or any other autonomous rideshare service but NOT TESLA. Period.
Your algo did not recognize the parked motorcycle on right. Check the clip at exact 4:59 and 5:00, the motorcycle isn't highlighted. I can guess why this happened. Hope you guys will be aware of this.
I wonder if the image detection is trained only on bikes that have a rider. I imagine it's a fair choice to treat it otherwise as a static object, because unlike a car where you can't see the driver so well, there is little risk that it will suddenly swerve out from where it's parked.
It just make me realize how far this kind of autonomous driving is from being able to drive in Europe, where the lanes are narrower, and the stop lights are not on the other side of the intersection and very often occluded.
Well they have almost 3x as many sensors and cameras as the Tesla had plus a more powerful computer in the back processing all this. On top of it they are using 8 lidar sensors where Tesla had none. If Tesla put as many sensors on their car the fsd package would cost $50,000 instead of $8,000.
Review God USA This quality FSD will not happen any time soon. Wake up. This is just a movie. What you see is a lot of post processing to make an exiting movie. Well done still. All I am asking for is a live demo. Like all the Tesla live demo’s that you can find on UA-cam, demonstrating how tough this is.
@@angadsingh9314 Yes, you are right, I was quite wrong. The FSD beta demo’s of today are quite impressive. I can now see how autonomous driving could really happen in the future. I have seen many of the video’s. Again, I was quite wrong.
I heard multiple references to behaviour that depends on the behaviour of other vehicles. For example 17:10 traffic flow to determine if it's ok to go through a yellow light. Are you programming the car to drive off a cliff as long as everybody else is?
Probably not very well. "Rare objects" like skateboarders, dogs, deer, etc are rarely seen by the car in test drives, so there usually isn't enough data to learn how to detect and plan around these objects. Some teams are working on learning from simulation (which would help with this), but I think that technology is still in its early days
If you will be driving much in San Francisco, not smart to own an expensive shiny car since it will get dinged, plus the vandals on the street who scratch cars and smash windows. The biggest difficulty driving are all the pedestrians, many who virtually jump from the sidewalk in front of your car. There are strict pedestrian rights and high fines for being in a cross-walk with pedestrians, so that some seem to take advantage of their power under the law. You must be constantly vigilant and never even glance at a device while driving there, except at a stoplight. If it can navigate in San Francisco, it would have no issues anywhere else in the U.S., except maybe Boston.
Not using pre-generated maps is like cutting off your left foot - you can still walk but it'll be hard and unsafe. If you know where you expect your lane lines to be, you can avoid scenarios where lane lines are faded, painted over, or obscured. Similarly, knowing where buildings, traffic lights, and signs are "supposed to be" makes perception and planning much more reliable - it's similar to having another sensor input that you can check. All serious self driving car companies use such maps (Waymo, Zoox, Cruise). Telsa doesn't use these maps, and you end up with shit like this: www.bbc.com/news/technology-51645566 where the car thought the "gore area" was a lane
Its not a game, its trying to achieve better safer way of transport and using predefined maps gives them an edge and they should use it. I also see ur point that its not scalable "easily" not sure how they expect to solve it
@@SydneyEV Roadworks are planned and documented. I like this solution because it improves safety. Of course having the system also be reliable without a map is nice, but this just gives you more trust in the system if you know there has been no work on that road.
Using a "pre-made map" GPS when you drive your car is also cheating. As is using a paper map. Real men don't use maps, they just guess where they're going and hope it's right.
not frustrated with zoox vehicle speed, there was another vehicle stopped in front of the zoox car, so the driver behind the zoox car went around, which the zoox car then also did. the driver behind was just more aggressive than the zoox car was in that situation.
Commercially unavailable and commercially unviable. But great tec for $900m investment. Oh, and you need a safety driver too. Seems this self driving idea just don't work.
Why would Amazon buy it if it is unviable? Most likely not viable for consumer cars, but I'm pretty sure Bezos has some good ideas where to use this system when he spends more than a billion dollars.
That's actually just driving over a very uneven manhole cover where the ground is bumpy. If you look at 26:21 you can see it clearly in front of the vehicle on the right side.
@@idaho5k736 I misread your comment but I think they meant FedEx and UPS, just throw a driver behind the wheel and they'll expect him to do twice the work lol
This is bullshit. They render the scenes multiple times. When you see 10+ vehicles off in the distance already identified before us as a human can even see them, you know it's time to call bullshit. Show us some real time rendering. Allow us to upload our own driving videos and see it in real time. Until then, stop wasting peoples time.
I'm an OpenPilot user and a software engineer. This system blows my mind.
What system? You are only seeing what they want to show you. The videos are pre-rendered multiple times with human interaction to identify things.
Yeah but NOBODY can buy a Zoox
car tomorrow, while anybody could buy a new car today and run OpenPilot on it immediately.
@@user-lt6wp4yd8y Unless they are lying about the car driving itself, I couldn't care less. This is impressive.
@@catbert7 ki 🙏
@menachem 5780 Tesla's systems are nowhere near as complex!
That was utterly impressive! I say this as an enthousiastic Tesla (FSD) owner.
Yes. I really wonder how good Tesla's system is in their engineering builds?
@menachem 5780 Zoox was founded in 2014 by Australian designer Tim Kentley-Klay, and Jesse Levinson, who was developing self-driving technology at Stanford: en.wikipedia.org/wiki/Zoox_(company).
The Tesla lawsuit alleged theft of line drawings of a warehouse layout, and had nothing to do with self driving, software, or Autopilot: www.theverge.com/2020/4/15/21222763/tesla-zoox-startup-lawsuit-settlement-autonomous-driving
Seeing all the complexity that ZOOX can navigate through is very impressive, and shows the importance of using lidar, radar, camera and mapping, and having the 360 degree view.
Roger Zimmerman: Yes. It's just ludicrous how Musk just denies any of the multi-sensing modes matter, AND detailed HD mapping (which Mobileye has a brilliant solution for, re crowdsourced updating with a narrow bandwidth data stream to keep the maps fresh), and the Tesla fanbois just blindly lap it up, even though Tesla is WAY behind in complex situations.
What's really going on re real world performance, often illustrated by UA-cam videos, tells the real tale.
Hi everyone. Im really interested why the stationary car have grey smaller boxes inside, and moving don't? These grey boxes come probably from lidar - are they just filtered out for moving object for rendering? Or there is a reason for that?
Fascinating. Would be great to see the behavior prediction for some agents!
So since this system is dependent on HD maps it can't be depolyed to any area that has not been mapped? What happens if construction changes the map?
Samuel Obirieze: Watch videos on Mobileye or read about Mobileye. They have a crowdsourced solution, which provides realtime updates. It's so good they have a business where they sell data to various cities, and plan to expand that massively.
It's brilliant, and makes the usual Tesla claim that HD maps are a problem a thing of the past.
@@rogergeyer9851 very unscalable solution. how do you map every single city in the world and not spend shit tonnes of money?
instead of red light signal, if trafic police will stop the line from (police standing far away), will the car stop?
Karmesh Madhavi: Per the video, the car understands gestures. It stopped for a pedestrian when the pedestrian held his hand up. No reason it couldn't stop for police. And given how poorly many people drive and don't pay attention, I would trust that system by the time it's ready for real robo-taxis with no safety driver far more than I would the average driver.
Whats presented here is fantastic. Two more steps needed here. First, add the steering wheel video to a corner. Second, do it live.
Alper ALT: Re the driving, they ARE doing it live. And it keeps getting better.
Re demo videos, what does it matter if they enhance the video later to make it easier for people to understand? What matters is that the CAR can understand things and handle them properly in real time.
This is light years ahead of Tesla, given what UA-cam shows Tesla doing in complex city traffic situations.
Compared to situations like what Zoox and Mobileye are routinely handling, Tesla is FAR away from even level 4 (at a reliable taxi level) for complex city driving, regardless of what Musk may claim.
Waymo One is harder to gauge. They can do level 4 for a small and simple scenario by avoiding the more complex routes and behaviors (like difficult left turns). So unless Waymo One has lots going on they aren't showing, I think they're well behind.
@@rogergeyer9851 I was not aware they were broadcasting these driving events, live. Tesla is comparable to Openpilot I believe. Both using cameras and radar, only. Zoox is doing everything right. And I don't know about mobileye's progress.
On 21:49 there are grey blocks in the middle of the road, that probably comes from traffic lights (over the road) - and they again are rendered on the ground level). These appear just for one cycle - interesting, if they have problems with underdrivable objects.
phenomenal. great stack!
Cruise cameo at 8:50
That's an interesting catch!
Awesome
This machine learning technology is amazing. Still not sure about how cost efficient that lidar & everything is. Tough to mass produce. seems very precise. Mapping streets seems life a flaw in teaching the AI to respond to unique scenarios. Roads are always changing
Mayko Marquez: But in the real world, there are solutions. Like the Mobileye crowdsourced, real time updates to its HD maps, which also gives it lots of data about city status (accidents, water leaks, fires, etc) which it is selling to cities for revenue -- plus the maps help the car drive much better.
Musk just ignores all that, claims vision will solve everything, expects everyone to believe him, and is DEAD WRONG, re what happens vs. his predictions, every single year, re autonomous driving.
I think pretending like Tesla doesn't have VERY serious competition and in fact that Tesla isn't seriously behind for complex driving, is just delusional. But of course, Musk won't admit this as it would likely significantly hurt Tesla's stock price.
But over time, the real world will demonstrate what's up with working robo-taxis at level 4 from the likes of Mobileye, Zoox, and possibly Waymo, and Tesla will always be "working on it", unless Musk and team wakes up to the reality.
How much of this city is coded for in the system? That is, can this car be dropped in New York with similar performance?
Wow 6 yrs of process really shows the progress & I'm seeing the only downside is the premapping of the area/city.
TESLA ARE YOU SEEING THIS.
WizNitro 07: Plus, the premapping has a lot of serious advantages. Like better understanding all the traffic lights in complex situations and which apply to which roads.
And, Mobileye has a very cool solution for the need for HD maps, updating the maps dynamically through crowdsourced cars, with very low bandwidth data streams. So it's really not much of a disadvantage to do the mapping if done right, AND all the advantages make the driving much more accurate. In a video like this, where the streets are crowded and narrow, clearly accuracy is ABSOLUTELY essential to safe driving over time.
Seeing videos like this and like what Mobileye is doing, and comparing them to the chaos we see from Tesla in complex city situaitons (where FREQUENT driver intervention is required, and a frightening number of times, the car would cause a crash or even bad crash without that intervention, tells me that Tesla is WAY behind on this vs. the leaders, regardless of what Musk (and the Tesla fanbois) are claiming.
And Tesla might get to the level 5 ungeofenced solution some day, but it sure as he** won't be 2021, and I don't see it happening before about 2030 on a truly global level. Musk should smarten up and be honest and look at a city-at-a-time solution for robo-taxis. Of course, to do that, he'd have to ADMIT that Tesla will have LOTS of city by city competition, and there goes the assumptions about unending massive Tesla income from high robo-taxi revenue, which will hit TSLA. So there's that.
At 12:52 many traffic lights are displayed, even though they are out of sight. Are they pre-mapped or how does the car know that they are there?
They're carefully pre-mapped (though as explained at 18:30 they rely on cameras to detect and reason about the current state of the lights). Traffic lights don't change often, figuring out the association between light and lane in real time is nontrivial to do 100%, and you *really* don't want to hit edge cases like twitter.com/derekteay/status/1244450876774907904.
Thanks for the explanation. Somehow I missed that part. Would be interesting to know how they handle temporary traffic lights at construction sites.
24:55 why did you take that left turn despite the no left turn sign?
That sign limits left turns to only be permissible at certain times of day. The left turn was legal when the drive happened.
@@TaylorA: And if the car can do things like that reliably (helped by the HD mapping, no doubt), it's a better driver than MANY drivers.
I've missed such signs driving in a strange situation the first time for example. And of course, see MANY, MANY people ignoring or missing such signs -- even the simpler ones that just say something like "no turn on red".
At 18:39 the second object behind ego car is extremely unstable - I understand, that it is occluded for almost all sensors, but can be that this instability is caused by interplay of camera detection and ML based tracking? Pretty interesting effect.
Hm, on 22:36 it is seen on the object on left lane that it is "stuck" for couple of cycles - probably there is no tracking it all? Strange.
Great stuff; I'm also interested how this would function at 10 PM at night, where camera (i.e., visual) data is less reliable and radar and LiDAR is more important.
How do you address exceptions like giant potholes without swerving onto oncoming traffic or hitting a pedestrian?
The same way a human driver does.
It took a while for me to see the 2x speed. I was like, it’s driving waay too fast, lol
This is amazing
Not really, using HD maps obviously
Amazon acquires you guys. Bravo!
would be interesting to see how Zoox managed flight risk in a tight street such as around lombard street. How does it handle flight risk from very young children or small pets such as dogs? this is currently a headache for humans, requiring 100% alertness and immediate response.
Henry Lim: If you watched and listened to the whole video, by implication, they made it quite clear. Planning, wide field of view, LOTS of sensors, and being damn careful.
Pretty much the opposite of Tesla in its current state, BTW.
So does the AI know if an emergency vehicle is nearby and how to react ?
@@hypercept forget dispatch in the future the emergency vehicle will signal the cars. Probably will signal traffic lights too.
congratz on amazon purchase
Very awesome!
What happens if your car break down on slope?
It will be interesting to see what fail safe is in place for that.
20:18-20:23 when you see 3 double parked vehicle on left lane, why did you still try to go left between 2 of them and then go right to bypass?
they just have to teach the software through similar experiences & take proper action. Some situation they didn't account for?
6:48 that's got to be a TikTok being made on the left
Amazing♥
very interesting
Funny, that when some part of pedestrian is occluded, the boxes on abstract visualization becomes shorter, but still are placed on the ground level - I know, it is just visualization, but rises questions about quality in other areas. ;-)
Please start Zoox taxi service in Scotland.
I can't stand not having the app to hail my Zoox ride.
I want to stop giving my money to mafia run taxi services and mafia run bus drivers.
Come on Zoox we know you can do it.
kind regards
LM
p.s. Would a city full of these vehicles emitting radar harm the surrounding environment?
Example, would a flower sat in a pot at the side of the road on a shelf survive its usual season,
or would the fine leaves weather due to exposure to these radar emissions?
Flowers never lie, they either survive or don't.
What vehicle are they using? Tesla? Waymo? And is it purely vision or is there LiDAR involved?
It's a Toyota Highlander I believe. They talk about your last question in the video quite extensively, I recommend watching it
Drive In Mumbai,India. I will agree to that only.
Imagine a world where autonomous cars had always existed and now people were arguing that humans should now also be allowed to drive. I think if you showed them this video, people would call you insane, to imagine that humans could navigate this.
Manuel Barkhau that’s stupid lmao
This flexes so hard on every autonomous vehicle company
I dunno about that. Cruise has a video out that looks exactly like this. Waymo has had this for years. Tesla is just ramping up their solution, which doesn't require maps. What baffles me is what's taking these mapped companies so long to mass produce and start servicing if it's as good as they're showing here.
@@catbert7 Maybe capital?
@@angadsingh9314 You're saying Google doesn't have enough capital? ;p
Three Modalities - CAMERAS, RADARS AND LIDARS - STAGGERING 18 Cameras, 10 Radars AND 8 LIDARS - "This many sensors will be pretty expensive and UGLY if you had to pay for them on your own car" They are competitors for Waymo/uber or any other autonomous rideshare service but NOT TESLA. Period.
Your algo did not recognize the parked motorcycle on right.
Check the clip at exact 4:59 and 5:00, the motorcycle isn't highlighted.
I can guess why this happened. Hope you guys will be aware of this.
@misterioszg true, I missed it.
I wonder if the image detection is trained only on bikes that have a rider. I imagine it's a fair choice to treat it otherwise as a static object, because unlike a car where you can't see the driver so well, there is little risk that it will suddenly swerve out from where it's parked.
It seems too clean . What is the difficulty that they having issues is not state at all so I can't judge on only one side of things.
It just make me realize how far this kind of autonomous driving is from being able to drive in Europe, where the lanes are narrower, and the stop lights are not on the other side of the intersection and very often occluded.
Tesla will be trying to be more inclusive
Also we nee to see accidents perdition also its mistake and error.
This system is unbelievable. Please help me believe it and do a live demo. I used to believe the Tesla video. Not any more.
Well they have almost 3x as many sensors and cameras as the Tesla had plus a more powerful computer in the back processing all this. On top of it they are using 8 lidar sensors where Tesla had none. If Tesla put as many sensors on their car the fsd package would cost $50,000 instead of $8,000.
Review God USA This quality FSD will not happen any time soon. Wake up. This is just a movie. What you see is a lot of post processing to make an exiting movie. Well done still. All I am asking for is a live demo. Like all the Tesla live demo’s that you can find on UA-cam, demonstrating how tough this is.
@@theoschijf8155 Are you sure? Nowadays, Tesla's FSD is near perfect in California.
@@angadsingh9314 Yes, you are right, I was quite wrong. The FSD beta demo’s of today are quite impressive. I can now see how autonomous driving could really happen in the future. I have seen many of the video’s. Again, I was quite wrong.
I heard multiple references to behaviour that depends on the behaviour of other vehicles. For example 17:10 traffic flow to determine if it's ok to go through a yellow light. Are you programming the car to drive off a cliff as long as everybody else is?
Lmao 🤣
I wonder how Zooks cars would respond to skateboarders.
Probably not very well. "Rare objects" like skateboarders, dogs, deer, etc are rarely seen by the car in test drives, so there usually isn't enough data to learn how to detect and plan around these objects. Some teams are working on learning from simulation (which would help with this), but I think that technology is still in its early days
Are y’all hiring ???
If you will be driving much in San Francisco, not smart to own an expensive shiny car since it will get dinged, plus the vandals on the street who scratch cars and smash windows. The biggest difficulty driving are all the pedestrians, many who virtually jump from the sidewalk in front of your car. There are strict pedestrian rights and high fines for being in a cross-walk with pedestrians, so that some seem to take advantage of their power under the law. You must be constantly vigilant and never even glance at a device while driving there, except at a stoplight. If it can navigate in San Francisco, it would have no issues anywhere else in the U.S., except maybe Boston.
Yeah but you have already scanned every road beforehand right? That's cheating
Not using pre-generated maps is like cutting off your left foot - you can still walk but it'll be hard and unsafe. If you know where you expect your lane lines to be, you can avoid scenarios where lane lines are faded, painted over, or obscured. Similarly, knowing where buildings, traffic lights, and signs are "supposed to be" makes perception and planning much more reliable - it's similar to having another sensor input that you can check.
All serious self driving car companies use such maps (Waymo, Zoox, Cruise). Telsa doesn't use these maps, and you end up with shit like this: www.bbc.com/news/technology-51645566 where the car thought the "gore area" was a lane
Its not a game, its trying to achieve better safer way of transport and using predefined maps gives them an edge and they should use it. I also see ur point that its not scalable "easily" not sure how they expect to solve it
Good point but hey if it works after scanning the roads and if they can afford that expense and accept the limitations, then why not?
@@SydneyEV Roadworks are planned and documented. I like this solution because it improves safety. Of course having the system also be reliable without a map is nice, but this just gives you more trust in the system if you know there has been no work on that road.
Using a "pre-made map" GPS when you drive your car is also cheating. As is using a paper map. Real men don't use maps, they just guess where they're going and hope it's right.
zoox vs zoom
7:04 driver behind is frustrated at the speed.
not frustrated with zoox vehicle speed, there was another vehicle stopped in front of the zoox car, so the driver behind the zoox car went around, which the zoox car then also did. the driver behind was just more aggressive than the zoox car was in that situation.
Commercially unavailable and commercially unviable.
But great tec for $900m investment.
Oh, and you need a safety driver too.
Seems this self driving idea just don't work.
Tesla is making it work, so far they're the only one
Why would Amazon buy it if it is unviable? Most likely not viable for consumer cars, but I'm pretty sure Bezos has some good ideas where to use this system when he spends more than a billion dollars.
26:15 Kissing the curb much.
good catch lol
That's actually just driving over a very uneven manhole cover where the ground is bumpy. If you look at 26:21 you can see it clearly in front of the vehicle on the right side.
@@TaylorA well spotted.
no view of the driver seat, i call bullshit.
This has the lo potential to team with Amazon and out a lot of delivery people out of business. Yikes.
Amazon just bought them
@@stevensonjuillet6835 thanks for the info. Still don't see how it is going to deliver to a doorstep with nobody home. Any info on that.
@@idaho5k736 I misread your comment but I think they meant FedEx and UPS, just throw a driver behind the wheel and they'll expect him to do twice the work lol
Then invention of the car put a lot of farriers and blacksmiths out of business ;)
This is bullshit. They render the scenes multiple times. When you see 10+ vehicles off in the distance already identified before us as a human can even see them, you know it's time to call bullshit. Show us some real time rendering. Allow us to upload our own driving videos and see it in real time. Until then, stop wasting peoples time.
I mean.... humans don't have perfect eyes...
It is terrible to imagine what will happen if this or a similar system is hacked by hackers. And there are reasons for this.
Wow the chic is fine!
Shame in you team, you have stolen secrets from Tesla
There's no secrets to this. It's been a LONG road to get many many different hardware and software aspects to this point. Decades.
Backwards thinking 20th century people are everywhere..