3 ms·

Is this how we finally get full self-driving? Maybe have enough compute in the car to take care of mundane cruising and emergency braking, checking with a centr

Is this how we finally get full self-driving? Maybe have enough compute in the car to take care of mundane cruising and emergency braking, checking with a centralized/shared expert-driver AI in exceptional circumstances. Maybe the expert-driver AI contacts an even-more-expert AI in even-more-exceptional circumstances. Etc. AIs all the way up. Imagine the implications.

inductive_magic 3y ago
>Imagine the implications. > exceptional circumstance appears > no network > implications
otabdeveloper4 3y ago
We will never have self-driving. Any attempt to sell this technology is a pump-and-dump scam. We have had fully autonomous airplanes and trains for decades now. The technology has been shelved because of legal reasons, not because it doesn't work. Self-driving technology will only last until the first child gets run over by a self-driving car, at which point the technology will be banned. Lawyers will need a fall guy to take the blame, and since tech bro corporations have a EULA that absolves them of all responsibility that will be the end of self-driving.
- jeroenhd 3y ago
  If companies take responsibility for their technology, the technology is viable. Various low speed self driving vehicles are already deployed all around the world. Low speed to reduce inevitable injury, which keeps risk to the company behind them down. The problem with self driving technology is that companies like Tesla want to keep you liable for the decisions their AI makes. That's why you have to sit behind the wheel and still keep an eye out. Here in the Netherlands there were several tests with self driving trains, but the systems didn't prove satisfactory yet. The systems are being worked on, though. Putting passengers in the hands of automated systems requires a lot of care and responsibility, and a few companies are willing to take that risk (for plenty of money, of course). As for planes, autopilot is used very often. Automated landing is mandatory in some runways to keep time between planes down, though pilots don't seem to like the aggressive touchdown these systems provide. We'll never have fully autonomous planes because someone needs to be around when the plane runs into trouble; computers won't land you on the Hudson.
  - 7952 3y ago
    Most pilots are probably not going to land you on the Hudson either. A computer designed to perform emergency landings may actually do better. It can instantly decide if a high bank angle turn will work. It can carry mapping data to identify possible landing locations and places to avoid crashing (like areas with higher population density). It could design an approach and fly it using GPS. It could use transponder data to land near vessels that could assist. You could even have published procedures for dual engine failure. In certain situations humans do not respond in an ideal way. They are bad at deadling with conflicting information. And will often make poor choices when faced with an emergency. A suitably designed automation may well do better in those situations. And a lot of accidents now are exactly those kind of situations. They are survivable incidents where the pilot fails to take the correct action. I don't think we should give emergency control to autopilots just yet. But the scientific problem it poses is worth tackling. How do you deal with possibly conflicting information and be sure that the reported situation and plan is correct. That is the problem that underlies a lot of challenges (like ATC, IFR navigation, search and rescue, routing drones, even command during a war). You have a set of data from the machine, recorded data about the external world, data from other machines and control, rules about expected behaviour, mapping data, model outputs, weather information. And that all has to be tested, fused together and then acted upon.
- kakwa_ 3y ago
  > We have had fully autonomous airplanes and trains for decades now. On the surface, yes. Looking more closely, kind-of-but-not-really. A plane auto-pilot is not really meant to fly the plane from gate to gate. It's more a series of automation simplifying each action performed during the flight. But each action have to be triggered individually and controlled by the pilots. For trains, full automation is not common at all. You see it see it mostly on a few metro lines, which is an ideal case (platforms can be secured, homogeneous equipment, very static route, short track enclosed in tunnels). On a classic railway networks, we are still at the stage of a few experimental projects. So, even with trains and planes, we are hardly there yet. And these are much simpler systems to automate: they are operating in heavily controlled environments, with a traffic that is not that dense (compared to a city street) and with tons of help from the infrastructure (think ILS for plane landings, or ETCS signaling & speed control for trains). So, a self-driving car in an open environment like a street? No, we are there yet, and in my opinion, we are still one or two decades away from seeing that.

inductive_magic 3y ago

>Imagine the implications. > exceptional circumstance appears > no network > implications

otabdeveloper4 3y ago

We will never have self-driving. Any attempt to sell this technology is a pump-and-dump scam. We have had fully autonomous airplanes and trains for decades now. The technology has been shelved because of legal reasons, not because it doesn't work. Self-driving technology will only last until the first child gets run over by a self-driving car, at which point the technology will be banned. Lawyers will need a fall guy to take the blame, and since tech bro corporations have a EULA that absolves them of all responsibility that will be the end of self-driving.

jeroenhd 3y ago
If companies take responsibility for their technology, the technology is viable. Various low speed self driving vehicles are already deployed all around the world. Low speed to reduce inevitable injury, which keeps risk to the company behind them down. The problem with self driving technology is that companies like Tesla want to keep you liable for the decisions their AI makes. That's why you have to sit behind the wheel and still keep an eye out. Here in the Netherlands there were several tests with self driving trains, but the systems didn't prove satisfactory yet. The systems are being worked on, though. Putting passengers in the hands of automated systems requires a lot of care and responsibility, and a few companies are willing to take that risk (for plenty of money, of course). As for planes, autopilot is used very often. Automated landing is mandatory in some runways to keep time between planes down, though pilots don't seem to like the aggressive touchdown these systems provide. We'll never have fully autonomous planes because someone needs to be around when the plane runs into trouble; computers won't land you on the Hudson.
- 7952 3y ago
  Most pilots are probably not going to land you on the Hudson either. A computer designed to perform emergency landings may actually do better. It can instantly decide if a high bank angle turn will work. It can carry mapping data to identify possible landing locations and places to avoid crashing (like areas with higher population density). It could design an approach and fly it using GPS. It could use transponder data to land near vessels that could assist. You could even have published procedures for dual engine failure. In certain situations humans do not respond in an ideal way. They are bad at deadling with conflicting information. And will often make poor choices when faced with an emergency. A suitably designed automation may well do better in those situations. And a lot of accidents now are exactly those kind of situations. They are survivable incidents where the pilot fails to take the correct action. I don't think we should give emergency control to autopilots just yet. But the scientific problem it poses is worth tackling. How do you deal with possibly conflicting information and be sure that the reported situation and plan is correct. That is the problem that underlies a lot of challenges (like ATC, IFR navigation, search and rescue, routing drones, even command during a war). You have a set of data from the machine, recorded data about the external world, data from other machines and control, rules about expected behaviour, mapping data, model outputs, weather information. And that all has to be tested, fused together and then acted upon.
kakwa_ 3y ago
> We have had fully autonomous airplanes and trains for decades now. On the surface, yes. Looking more closely, kind-of-but-not-really. A plane auto-pilot is not really meant to fly the plane from gate to gate. It's more a series of automation simplifying each action performed during the flight. But each action have to be triggered individually and controlled by the pilots. For trains, full automation is not common at all. You see it see it mostly on a few metro lines, which is an ideal case (platforms can be secured, homogeneous equipment, very static route, short track enclosed in tunnels). On a classic railway networks, we are still at the stage of a few experimental projects. So, even with trains and planes, we are hardly there yet. And these are much simpler systems to automate: they are operating in heavily controlled environments, with a traffic that is not that dense (compared to a city street) and with tons of help from the infrastructure (think ILS for plane landings, or ETCS signaling & speed control for trains). So, a self-driving car in an open environment like a street? No, we are there yet, and in my opinion, we are still one or two decades away from seeing that.