Self driving buses, trains, trucks etc

magicbastarder wrote: »

Bikes put spanner in works of Dutch driverless car schemes
Report highlights problems bicycles cause to self-driving cars’ detection systems
https://www.theguardian.com/world/2019/feb/13/bikes-put-spanner-in-works-of-dutch-driverless-car-schemes

I think that we're many years away from trusting autonomous vehicles on a shared use road, on motorways or other powered vehicles only roads, then only a few years.

is_that_so wrote: »

It's not an idiot designer issue. Cyclists, like pedestrians, are an unpredictable variable hence the suggestions in the article to explore larger road options first.

It is there variables that will ultimately stop autonomous vehicles from ever using mixed use routes.

cnocbui wrote: »

I don't think it's so much the bikes putting a spanner in it so much as the complexity and skill set required to do the task of driving has been grossly underestimated and that the general confidence in the ability to create systems that can do the task has been severely misplaced.

You can't design for bikes because the task of driving, IMO, requires general intelligence, so it's a task that is too complex for our current level of technology.

For mixed use routes, I would agree, motorways on the other hand require a much lower level of "intelligence" to programme an autonomous vehicle to drive safely.

AndrewJRenko wrote: »

I'm not seeing why bikes would be a particular challenge, relative to other cars or pedestrians.





No, but they do tend to make fast, dangerous manoeuvres without looking or indicating while scrolling through Instagram. Why would there be a particular difficulty with cyclists, relative to motorists and pedestrians?

The active sensors on the autonomous vehicles need to be far more sensitive to detect and identify pedestrians & cyclists than ones that only need to identify other vehicles.

It's asking too much of the technology to make split second decisions and then change the direction of something that's probably moving at 100Kph and weighing 1-2 tonnes, based of the interpretation of what a small object is that is moving across the road 100metres ahead.

It could be a crisp packet or a cyclist, if it's a cyclist will it be safely across before reaching it, these thought processes are natural to a human, not to a computer. So my belief that autonomous vehicles will be restricted to motorways (with active transponders) will be as far as it will get for the foreseeable future is technically sound based on the abilities of AI today.

In reality, the roads would need to have compatible transponders that communicate the local topology to the vehicles onboard computers as opposed to relying on the vehicles sensors to make sense of the world around.

AndrewJRenko wrote: »

Maybe, but I'd have thought that cyclists would largely fall somewhere between motorists and pedestrians on all of those attributes, so I'm wondering why they are set up as 'the problem'.

The "problem" is for the system to keep track of multiple small objects that are detected by the sensors, identify them, predict their future motion and whether your presence and direction of motion will influence their future movements and how these movements will influence the movement of vehicles around you.

These are things humans are able to do almost without thinking.

prinzeugen wrote: »

Exactly. All these claims that AI and self driving trains, cars etc would be safer is bollox.

And they are programmed by humans. The X-Files did a good episode on it.

Self driving trains are real and there are several driverless rail networks around the world, in controlled environments where the "system" is separated from humans, autonomous systems are perfectly safe.

Flying is another form of transport that can be automated without too much difficulty, the pilot is totally dependent on the instruments anyway.