Is lidar the best sensor for autonomous driving cars?

[stickygalaxy]

If we can have very good autonomous driving, we can reduce insurance costs by 10 times if there are 10 times less collisions. Isn't Lidar is a blunt instrument to use for autonomous cars? The billions of dollars of investment in lidar are not expected to bring hands free commuting in any know timeframe.

Radios can transmit displacement vectors, velocity, inertia, proximity, chassis limits, sudden braking, abs status and wheel turns of nearby cars, both to a dashboard radar display and to the car safety calibration system and auto nav system. Why use lidar?

It would make sense to have a $50 radio transmitter in every car as a sensor for the car computers, rather than a $200 lidar, because it would reduce collisions by 90%.

Future Technology Blog: ReCal radio vision autonomous cars divide insurance costs by 10

 

[Barklikeadog]

It is hard to believe that we can have a successful completely autonomous car until all of the other vehicles on the road communicate with each other. And the average car on the road is 12-15 years old, so there's that.

 

[djgarrett21]

It is hard to believe that we can have a successful completely autonomous car until all of the other vehicles on the road communicate with each other. And the average car on the road is 12-15 years old, so there's that.

I'm of the opposite mind. I believe relying on communication between autonomous cars is a means of defeating autonomy itself; vehicles become dependent upon each other. A sensor error in one vehicle can have catastrophic consequences for another. Also that type of network can/will inevitably be compromised creating a single point of failure for all vehicles.

My feeling is that true autonomy will require vehicles to have enough intelligence to make driving decisions independently of a network or other vehicles.

That's why I think Tesla's approach to "general autonomy" is the right one. Minimizing crutches and failure modes will win out in the long run.

 

[ewoodrick]

Radios and other car communication have a few issue with hacking.

Humans have shown that cars can be driven safely with a single stereoscopic camera that can be pivoted.

Humans have also shown that if you stop paying attention to the road, bad things can happen. In very, very few situation have attentive drivers had issues.

 

[electracity]

How is your radio equipped cars avoiding non-cars?

I think it is becoming clear that car will have radios, but also function independently.

Gas cars are already on a schedule to be banned from city centers. The second wave may be to ban human drivers from city centers. Those private autonomous vehicles will probably be required to cooperate with the transportation network. To meet that requirement the vehicles would have a radio.

 

[stickygalaxy]

The ideal is to reach a zero road death level and automatic commuting. So all the conceivable sensors and emitters have to be added to the car. Radio transmission and phone signal trackers and lidar, that's three sensor types. I would go 140 on the freeway if every car had 2 radios, a $10 12v radios on the battery sending out vehicle size and radar pings, and a 30 dollar radio connected to the onboard computer. You'd have to switch off autonomous mode if there was a radio error. oh no! Lifetime guarantee electronics and Toshiba are not the same.

I checked the figures for pedestrian accidents, they account for about 10% of accidents and 10-30% of road fatalities.
Lidar and stereo cameras are the best for pedestrians, and if there are autonomous cars, it won't be safe just crossing the road with hands down, people will have to wave with major arm movements so that the cameras see them in time. Radars can sense when they are driving towards a phone emitter, and car AI can use phone positions as an added failsafe to reach the zero.

If we want to go at 140mph on the freeway, radios can do it For radio's there are backup radio's so it isn't only one of them. There's the failsafe one, which is as reliable as the battery, just on 12v, and there is a backup battery.

 

[Barklikeadog]

I'm of the opposite mind. I believe relying on communication between autonomous cars is a means of defeating autonomy itself; vehicles become dependent upon each other. A sensor error in one vehicle can have catastrophic consequences for another. Also that type of network can/will inevitably be compromised creating a single point of failure for all vehicles.

My feeling is that true autonomy will require vehicles to have enough intelligence to make driving decisions independently of a network or other vehicles.

That's why I think Tesla's approach to "general autonomy" is the right one. Minimizing crutches and failure modes will win out in the long run.

One of the most valuable and least considered aspects of human driving is seeing 5, 10, 20 cars ahead and knowing what is going on. on my daily commute, there is an off/on ramp clover with about 100 yards of straightaway. All of the cars getting off are immediately changing lanes left, all of us getting on are switching lanes right. It is a constant crisscross and the only way to pick your spot is to see the spacing of the cars coming around the raised loop. a car could have 50 cameras, it's not going to figure this one out unless it knows what the other cars are doing.

 

[bxr140]

One of the most valuable and least considered aspects of human driving is seeing 5, 10, 20 cars ahead and knowing what is going on.

Certainly augmenting autonomy will make for more comfortable and more efficient travels, especially congested ones like the example where slowing sooner achieves both a more comfortable and efficient journey. For all situations, especially congested ones, safety must be exclusively autonomous.

 

[Barklikeadog]

The ideal is to reach a zero road death level and automatic commuting. So all the conceivable sensors and emitters have to be added to the car. Radio transmission and phone signal trackers and lidar, that's three sensor types. I would go 140 on the freeway if every car had 2 radios, a $10 12v radios on the battery sending out vehicle size and radar pings, and a 30 dollar radio connected to the onboard computer. You'd have to switch off autonomous mode if there was a radio error. oh no! Lifetime guarantee electronics and Toshiba are not the same.

I checked the figures for pedestrian accidents, they account for about 10% of accidents and 10-30% of road fatalities.
Lidar and stereo cameras are the best for pedestrians, and if there are autonomous cars, it won't be safe just crossing the road with hands down, people will have to wave with major arm movements so that the cameras see them in time. Radars can sense when they are driving towards a phone emitter, and car AI can use phone positions as an added failsafe to reach the zero.

If we want to go at 140mph on the freeway, radios can do it For radio's there are backup radio's so it isn't only one of them. There's the failsafe one, which is as reliable as the battery, just on 12v, and there is a backup battery.

My points on this matter isn't so much to do with safety. Naturally they will be more safe. It is the possibility of
creating impossible traffic jams in unnecessary places

 

[CarlK]

It is hard to believe that we can have a successful completely autonomous car until all of the other vehicles on the road communicate with each other. And the average car on the road is 12-15 years old, so there's that.

Totally agree. Some auto companies are chasing V2V or I2V solutions but they will never get there if that's what they are relying on for autonomous driving implementation. You never hear Elon mention that. Human drivers do not have radio transmitter or receiver in their brain. The goal is eventually having zero traffic death but FSD will be a worthwhile thing to do soon as it can meaningfully reduce chance of accidents. That is a relatively easy goal to achieve as human are terrible drivers.

 

[CarlK]

One of the most valuable and least considered aspects of human driving is seeing 5, 10, 20 cars ahead and knowing what is going on. on my daily commute, there is an off/on ramp clover with about 100 yards of straightaway. All of the cars getting off are immediately changing lanes left, all of us getting on are switching lanes right. It is a constant crisscross and the only way to pick your spot is to see the spacing of the cars coming around the raised loop. a car could have 50 cameras, it's not going to figure this one out unless it knows what the other cars are doing.

Cameras can track multiple objects simultaneously better than we could. Even that you feel this way human never "see" more than a few objects in one glance. You're actually scanning the field in quick succession to pick up objects you're interested and ignore others. Machines can do that much faster. They only need to "learn" what to pick up and what to ignore.

 

[Barklikeadog]

Cameras can track multiple objects simultaneously better than we could. Even that you feel this way human never "see" more than a few objects in one glance. You're actually scanning the field in quick succession to pick up objects you're interested and ignore others. Machines can do that much faster. They only need to "learn" what to pick up and what to ignore.

At what point can a machine see brake lights a half mile ahead and know to slow down? Or see a police car stopped in the right lane and know to get over well in advance. Will a car ever be able to see across a field and gauge how fast traffic is going in order to merge without hitting the brakes? How about another driver waving her hand? Maybe I lack the imagination of what a camera can accomplish, but I think all vehicles on the road communicating with each other is vital.

 

[CarlK]

At what point can a machine see brake lights a half mile ahead and know to slow down? Or see a police car stopped in the right lane and know to get over well in advance. Will a car ever be able to see across a field and gauge how fast traffic is going in order to merge without hitting the brakes? How about another driver waving her hand? Maybe I lack the imagination of what a camera can accomplish, but I think all vehicles on the road communicating with each other is vital.

In theory machines can learn to do any of those things you mentioned. They just need to go through the tedious learning process. You might say theory does not matter you still need to make it happen. The fact is theory does matter until you have a theory to prove otherwise. Elon, in his first principle way of thinking, always say he first need to get the physics down and see a clear path. Then he can work hard to face tough engineering challenges.

As for V2V it will never be there for FSD if it has to rely on it to work. Even if you can equip 50% of cars on the road with the ability what would you do when you meet the other 50%? Using the test above there is no clear path to get there. That's the reason why Tesla is not banking on it and others who are will be getting nowhere.