Driverless / self driving cars and cycle training / bikeability

It going to happen. cars without drivers.
What are the implications of this for the cycle training industry?

Level 1 skills will always be needed.
Core level 2 skills such as communication and eye contact, positioning to be seen by drivers and perhaps even awareness may be redundant...

What outcomes would be needed for sharing roads with robocars?

You may need to find out what the cars are sensing for. Afaik they monitor signalling and road positioning of other vehicles - and driverless cars are programmed to give wide berth to vehicles that
Change lane without indicating
Accelerate decelerate erratically
Adopt poor lane position
Cut into others braking distance

So try and find out what the driverless cars pick up about cyclists.

Core level 2 skills such as communication and eye contact, positioning to be seen by drivers and perhaps even awareness may be redundant...

Not sure. I think these skills will just need to be readjusted to focus more into using other features (such as the speed, braking, steering movements) that we also use in trying to predict driver behavior. A number of new "game strategies" too, I suspect, will develop as we come to be able to expect certain kinds of responses to certain kinds of stimulae, resp. car sensor activation.
The biggest problem I suspect facing us as cyclists will be lobby groups and legislators that might push for an exclusion of bicycles from the main roadways-- just as bicycles came to be excluded from many highways (recall it was originally bicycle groups that lobbied for the establishment of the US highway system) and horses from city roads (again many were originally designed for horse and carriages).

Surely there are too many permutations to consider before knowing more? How the syllabus needs to change will depend on the regulatory / legal requirements of the vehicles and their passengers.

Presumably these vehicles will need to be able to adapt to cyclists on quiet roads with no infrastructure. I can't think much would need to change; maybe children (and adults) will trust the robots more than the 'scary' drivers...

I want a jammer that makes all motor vehicles stop within 50m of me.

Then I can just ride round them while they sit there helpless in their obesity-ennabling devices.

That device is called central London

I'm here all week

Don't give up the day job.

Just modify the outcomes yeah?

You have a point. Giving up the day job may be the result of this tech. (in fact giving up the day job may be the result of teaching a new generation of cyclists who can teach their own kids to cycle robocars or no robocars)

I'm rather concerned that at present with drivers using sensors it's not on an MOT or any other legal requirement to maintain the sensors(I remember it mentioned with all that came out after the lady cycling was killed at Bank). Having a car that then drives itself off them sensors is worrying and as much as they are testing them they aren't abusing the cars for a few years or having others crash into them then doing bodge repairs or doing any other things you see daily on the road. How many brake lights do you see not working? think of each one as a sensor gone and the potental for harm is very high imo.

I'm also 99-100% sure whatever they are programmed to leave the factory with once a few months pass alternative remap or racing chip style alternatives will appear to make your self driving car accelerate faster and drive more aggressively taking more risks.

One thing I hope is that the rival brands don't get programmed differently so then whatever it emerges we might need to teach about them is then a 50:50 with one brands reacting well to something and the other very badly.

The biggest issue I can see is when moving trainees in a snake a self driving car won't have the human understanding that all the riders are together or be able to understand that only the front+back people signal, this could make them constantly attempt to overtake or do all sorts of weird and dangerous stuff as it's interpretation of the riders could be way off the mark.

I'm also 99-100% sure whatever they are programmed to leave the factory with once a few months pass alternative remap or racing chip style alternatives will appear to make your self driving car accelerate faster and drive more aggressively taking more risks.

This is really only as option as long as these are, like the Tesla, not "self-driving" vehicles as part of automomous fleets but "driver assist" sports cars. Liability increasingly is on the side of the car makers and, in turn, on the software-- why there are already moves to open source things. Looking at the financial damage to Samsung that a battery can make in a single model (Note 7) or to VW what a bit of sneaky software can create ($10 billion in brand value and as much as $18 billion in compensation and fines) we are no longer in the world of the Ford Corvair ("Unsafe at Any Speed: The Designed-In Dangers of the American Automobile", Ralph Nader, 1965) but litigation without bounds..

Sure there will, I suspect, be high-end sportscars (the niche currently filled by Koenigsegg, Lamborghini, Lykan etc.)-- as outgrowth of the Formula-E-- but I suspect that insurrance and liability will leave their home turf to places like Shanghai, Moscow, Dubai, Riyadh and less so in London, San Francisco, Los Angeles.. Doubt you will see any in Nottingham..

Sim Bamford founded our worker's cooperative CTUK. He now works for IniLabs on Artificial Intelligence and has done some work on driverless cars ( @Velocio @hippy )

On the question of cyclists behaviour in relation to them (and the future of training) he says:

In general it's good news for everyone - they're not going to get on the roads unless they have a significantly better safety record than human drivers. There was a recent crash - the first fatality, where the car couldn't see a white van against a white sky. Now everyone, if they didn't think so before, thinks that cars will need a combination of sensors - not just cameras but also radar and lidar. However to the extent that cameras are used, even if just as secondary mechanisms, it still makes sense for cyclists to do with their positioning what they can to improve visibility. Also with radar, a computer has to classify obstacles and being further away from the side of the road means less possibility for confusion. Good positioning, of course, still gives you more options if other road users make mistakes. So I don't really think anything should change in training. There might need to be lobbying to ensure that those programming driving strategies don't get into a race to see which model can go faster by being more aggressive (e.g. pushing through tighter gaps).

http://inilabs.com/videos/#driving

Also in this video, see the cyclist at 33 s:

http://inilabs.com/videos/#drivingalghero

Phone is rooted and overclocked running a custom firmware, pc is overclocked, just about everything we pass to people is soon after cracked/hacked open and modified to barely be recognised. I don't see why this shouldn't be extended to car software or why anyone would expect any different.

If the algorithm says only accelerate 60% and pass cyclists with 1.5m space it's only a matter of time before someone changes it to 100% and 0.5m. It's a rather endless list of ways you could modify them to drive differently.

Phone is rooted and overclocked running a custom firmware, pc is overclocked, just about everything we pass to people is soon after cracked/hacked open and modified to barely be recognised

If you root your phone and brick it then you just have a bricked phone. If you overlock your pc and your system overheads.. then you just have an overheated system and perhaps a voided warranty. If your crack, hack or open a product to do mods you, at worst, ruin the product-- perhaps, at worst, break a few laws (e.g. radio emission etc.)-- and nullify its warranty.. If you modify the certified software of a self-driving car you probably well have also nullified your insurrance coverage.

If the algorithm says only accelerate 60% and pass cyclists with 1.5m space it's only a matter of time before someone changes it to 100% and 0.5m. It's a rather endless list of ways you could modify them to drive differently.

"Fortunately" it does not work that way..
See, for example, https://images.nvidia.com/content/tegra/automotive/images/2016/solutions/pdf/end-to-end-dl-using-px.pdf

That's a great pdf on how it can recognise anything from a roads lane markings to a gravel track and that it will center in the lane or track. Has 0 about other roadusers? Somewhere be it by picture examples the car will aim to replicate or by hard numbers, the car will be driving off a logical list of commands to follow.

Is car in lane? does it look like the picture? is it 0.2m left/right of center

check sensor/camera

car isn't in lane/doesn't look like picture/isn't 0.2m left/right of center

adjust steering wheel left/right

return to top

I don't think invalidated insurance ever stopped anyone doing anything, overheating and stuff is a common concern for existing car software mods too.
http://www.autoexpress.co.uk/car-news/consumer-news/59931/watchdog-engine-remapping

I think it's going slightly off topic maybe more a discussion for the generic self driving cars topic.

Anyway imo the big one for cycle training is stuff like this, when a human driver sees this they usually understand the riders are together "oh look the local school has cycle training" but when a self driving car sees this what will it do? does it see them all as individuals and constantly attempt to get between riders? Also with the talk about self driving cars maybe responding to hand signals what happens when only the instructors signal? Would/should this impact how trainees are moved to drill sites?

How many cyclists ahead can a self driving car define/map, if you put one behind an ASL in clapham with 20-30 riders infront of it would it detect all of them? Will it become a think campaign "Think! don't ride in groups of more than 5"

when a human driver sees this they usually understand the riders are together "oh look the local school has cycle training" but when a self driving car sees this what will it do? does it see them all as individuals and constantly attempt to get between riders? Also with the talk about self driving cars maybe responding to hand signals what happens when only the instructors signal? Would/should this impact how trainees are moved to drill sites?

I'm sure this won't be an issue. I can imagine SDC's being used as support cars for TdF so programmed to pick up snakes and peletons of cyclists

Depending upon the solution one sees...

1 Attachment

I don't think invalidated insurance ever stopped anyone doing anything, overheating and stuff is a common concern for existing car software mods too.

The threat of significant liability and probable imprisonment should things go wrong will surely stop most..

https://twitter.com/kvogt/status/1385292636039176192?s=19