An IEEE article discusses problems that autonomous vehicle systems are having with recognising bicycles and predicting their movements.
The following includes a thought about autonomous vehicles.
Can we have a mature discussion about the future of urban transport?
By Alan Davies
Published by Crikey on 30-Jan-2017
Republished by InDaily on 31-Jan-2017
With 90% of motorised travel in capital cities currently undertaken by private transport it’s time for a grown-up assessment of where to go with urban transport policy.
Private transport continues to dominate motorised passenger travel. It currently accounts for 94% of total kilometres travelled in Adelaide, marginally ahead of Perth (93%), Brisbane (92%) and Melbourne (89%).
The dominance of private transport in Australian cities is in stark contrast with European cities. For example, private transport’s share of motorised travel in central Paris is only 25% while public transport’s is a whopping 75%.
Travel by private modes could increase significantly when autonomous vehicles establish a sizeable presence in the vehicle fleet; that’s because they make time spent travelling less “costly” by enabling passengers to engage in other activities while in-vehicle.
Fully electric vehicles could encourage more travel too, because of their lower fuel cost.
Can you send the car home again after it has dropped you in the city, to avoid car park fees? We could see a doubling of traffic.
Depends on where "home" is of course. There are plenty of empty streets 1km or less from the parklands where your own autonomous vehicle could lay up for the day, or cheap parking would be made available for this purpose. Most cars would of course be more like Uber and just find another customer.
Love the XKCD but autonomous vehicles will not only be able to recharge/refuel themselves but would be able to recognise a human passenger with the correct authority before driving off.
What we need is self-driving bicycles. ;-)
Brilliant. I love the way it picks ups the kids from school.
My grandfather said he had one of those self-driving horse things that walk him home while he was asleep. My grandfather was asleep I mean not the horse.
Some of those percentages are worrying:
Deep3DBox is among the best, yet it spots only 74 percent of bikes in the benchmarking test. And though it can orient over 88 percent of the cars in the test images, it scores just 59 percent for the bikes.
So how do they factor for a missed positive given some of these vehicles have been driving around and apparently doing not so bad - or did I miss something?
They don't talk about how the learning systems deal with motor-bikes which would some kind of mid-point analogue.
Maybe the fast ones are mistaken for motorbikes and the slow ones are mistaken for obstacles. You don't have to correctly identify something to decide not to collide with it.
What's the fuss? Aren't those figures somewhat better than the average human driver's responses? :)