Comment by FL33TW00D
10 hours ago
How can you say camera only navigation won’t work with such finality when humans manage just fine every day! You literally have an existence proof of it working.
10 hours ago
How can you say camera only navigation won’t work with such finality when humans manage just fine every day! You literally have an existence proof of it working.
It would be possible to build an ornithopter, evidenced by the existence of avians, but it turned out the easiest ways to make flying machines were otherwise.
I like the comparison, but with aviation on a fundamental level we made it simpler (removing actuation), not added more (senses we dont need)
What counts is the overall complexity, not the complexity of a single subsystem.
Using more senses allows simpler processing of the sensor data, especially when there is a requirement for high reliability, and at least until now this has demonstrated a simpler complete system.
Others in this subthread discussed the comparison of the complexity of different ways of achieving flight itself, but I think there is an interesting discussion in that... well... we do add senses we don't technically need to achieve stable flight (but are very useful for safe flight and have reduced the incidence of aviation incidents and accidents dramatically).
Whether it be altimeters based on radio[1] or air pressure[2], avoidance and surveillance systems that use radio waves to avoid collisions with other aircraft[3][4], airborne weather radars[5], sensors that measure angle of attack (AoA), GNSS location, attitude, etc, many aircraft (even unpowered gliders!) have some combination of special sensing systems that aren't strictly necessary to take off, fly to a destination, and land, even if some are required for what many would consider safe flight in some scenarios.
Many of these systems have redundancies built in in some form or another and many of these systems are even built into unmanned aerial systems (UASes) big and small.
[1]: https://en.wikipedia.org/wiki/Radar_altimeter
[2]: https://en.wikipedia.org/wiki/Pressure_altimeter
[3]: https://en.wikipedia.org/wiki/Traffic_collision_avoidance_sy...
[4]: https://en.wikipedia.org/wiki/Automatic_Dependent_Surveillan...
[5]: https://skybrary.aero/articles/weather-radar
I'm not sure I agree. I think just having wings that flex a bit is mechanically simpler than having an additional rotating propellor. After all, rotating axles are so hard to evolve they never almost never show up in nature at a macro scale. Sort of a perfect analogy to lidar actually. We create a new approach to solve the problem in a more efficient way, that evolution couldn't reach in billions of years
3 replies →
Jet engines do not strike me as being inherently simpler than muscles, not by a long shot.
They're still the best way we know of going about the business of building a flying machine, for various reasons.
3 replies →
I would posit that the human brain is complex, and adding senses is simpler than replicating an aspect of the mind more accurately.
> easiest
This is the keyword here, just because the other approach is harder does not mean it is impossible.
It's a decent gamble to try and do things the hard way if it is possible to be deployed on cheaper/smaller hardware (eg: no lidars, just cameras).
Is it still a decent gamble after you've been trying (and failing) for a decade, and numerous well funded competitors are going the easy way, and when there is huge upside to being first, and when the value of FSD easily covers the rapidly falling cost of LIDAR?
No. It's not a good idea. It's not a good gamble. It's stupid, and the engineers can see it's stupid. A lot of them have quit, reducing the very slim chances of it working even further.
2 replies →
Because FSD driving not navigation is going to be held (rightly) to a much higher standard than human driving.
Humans are fallible and we have other sensors, like hearing, or touch (through feedback on the steering wheel) that are also involved in driving.
We already have other sensors that are not vision that work with us when driving like ABS and electronic stability.
The other reason it's dumb is that adding LIDAR and proper sensor fusion makes things better and the cost of LIDAR is rapidly dropping as its installed across new fleets in CN and elsewhere.
Yeah and we should replace the wheels with legs. every other company disagrees with musk, putting alternate sensors on even low end cars.
Both the vision and cognition hardware in humans are vastly superior, and don't get me started on the software.
I never understood why they would choose to fight with "one hand behind your back". More sensors = more better
~1.2 million global deaths per year due to motor vehicle accidents say otherwise.
Actually, that's the standard we're all talking about. Nearly everyone is totally fine with human-caused traffic deaths. Nobody wants to ban human drivers at that rate of death.
But if FSD had the same rate, people would be losing it.
The safety record of humans is not so great. They tend to fail in snow, ice, fog, rain and at night. We should be aiming a little higher.
I don’t think it makes sense to limit yourself while you are still figuring out what really works. You should go with a maximum of sensors and once it works, you can see what can be left out.
Yeah, but even if the safety level was 10% better, let's say--nobody would accept that rate. It wouldn't get adopted, we wouldn't be happy to save those lives. People would be outraged.
I think it's got something to do with an innate belief to self-determination. It's fine if I make a mistake to kill myself, and it's not fine if someone else does. It's super not fine if someone dies at the hands of a rich person's technology. Outrage, lawsuits, "justice."
Eyes have higher dynamic range and eyes don't freeze below 0°C. You can also put on sunglasses for even more weather-related adjustments.
While I am in the camp that believes camera-only FSD won't succeed, your comment isn't entirely accurate.
CCD and CMOS sensors can easily work in sub-freezing temperatures with various kinds of heating. There are 10's of millions of surveillance cameras installed outdoors in sub-freezing climates that work fine.
Cameras also have moveable IR cut filters, which is analogous to your sunglasses example.
Human eyes do have greater dynamic range in the visible light spectrum, but solid state sensors can commonly interpret light above 1000nm, and of course you can do thermal/IR imagers to provide optical sensing of wavelengths outside of what a human can see.
Sensor technology relative to the human eye isn't what is holding FSD back.
We don’t drive with just our eyes, we also drive with our brains.
Humans have cameras (eyes) + AGI. Cars have to compensate with LiDARs and other sensors that humans don't.
Technology can't compete with how easy it is to make more human-based navigation devices ;-)
This is commonly said but trivially falsifiable — a blind human crosses the street better than a Tesla.
Eyesight isn’t the thing. Humans have a persistent mental model of the world, and of the physics that drive it. Our eyes only check in every now and then to keep our model up to date.
Our ears and sense of touch do a lot of work in walking and driving, too. Trying to narrow it all down to vision is silly.
Deaf people drive.
I knew a guy with no arms who drove with his prosthetic hooks. Of course he can feel vibrations and things through his ass, but so could the car if they wanted. Do they use accelerometer data? (I don't know the answer to that) Do they have ABS sensors that can detect wheel lockup/speed status? Because I don't.
I believe I can drive a car to the legal standard, remotely, with a good enough camera array.
Because we can't install lidars on our heads
I fall on my butt all the time.
Because I want better?