Bits of Books - Books by Title
Intelligent Cars and the Road Ahead
Hod Lipson and Melba Kurman
99% of driving is routine: AVs have to deal with the corner cases - the 1% of the time where driving gets "exciting".
Character in Ernest Hemingway's The Sun Also Rises asks "How do you go bankrupt?" and the answer is "Two ways: Gradually and then suddenly." AVs like that - the tech is getting better all the time; then bang! it'll hit a tipping point.
Road accidents kill 1.2 million people a year. About 183,000 people die from drug causes (incl prescription drugs). Yet there is a 'war on drugs'? "Today's cars are deadly overlords."
Rethink car design - much of the weight of modern car is due to safety features.
Some driving will be completely replaced. You won't need a one ton car to deliver a one-pound pizza - instead they will use drones, possibly built round an oven to complete cooking pizza en route.
If could get rid of need to accommodate parked cars, would change cities. Not only free up streets, but also liberate businesses from meeting zoning requirements. The downside of course, is the loss of parking revenue for the cities. There are other unexpected consequences from parking decisions. LA built a concert hall with 6 levels of underground parking. Patrons drive into the garage and take an elevator to the concert hall. They never set foot in the streets around, which are now deserted and scary. In contrast, SF's concert hall parking is down the street a bit; patrons walk to the venue, and the streets are bright and vibrant.
Car makers favour an incremental approach, bc obviously that keeps them in the driver's seat for longer. But that is probably unwise, given how quickly 'drivers' get bored and stop monitoring what the car is doing. Some want to keep the human in ultimate control, but trying to get humans and robots to share responsibility just doesn't work. It's one of those cases where the theory is correct (the best results from automation are often when a human has the final say) but wrong in practice. People trust automation very quickly once they see it's working perfectly. What happens with AVs is that the 'driver' starts doing something else, and if called upon in an emergency, takes far to long to understand the situation. Basically Asleep At The Wheel.
So Google has conclude that there is no middle ground - you have to go straight to fully autonomous vehicles. And take away the steering wheel.
All but one of Google's car accidents due to cars driving too conservatively, and having others drive into them. Short term solution will be to program cars to drive more like humans. (Then once all cars in fully AV situations, can apply different set of rules).
Earlier idea of intelligent highways much too expensive. Only 'improvement' AVs need are clear lane markers.
3 DARPA Challenges in 2004, 2005 and 2007. First offered a cash prize of $1 million for the winner of a 150 mile race through Mohave Desert. It was a fiasco. None of the 15 competitors got more than 8 miles from the start. The second race 5 vehicles managed to steer themselves around the course. There are 2 ways to program an AV. One is to try to anticipate everything that can happen, and to program a rule for that telling the vehicle what to do. The other way, which is what the winner used, was AI - basically a whole lot of trial-and-error to teach the robot brain how to drive.
Essentially, AI means teaching computer to make and improve predictions based on supplied data. Although we say "learning", what is actually happening is that an algorithm parses huge amounts of data to look for statistical patterns. It then builds a mathematical model that ranks the probability of various possible outcomes to make predictions or make a decision. The machine then tests its predictions on new data. If they are wrong it goes back to update the model.
This is how autocomplete works - it uses both general data and a record of what you've typed in the past to make prediction, based on statistics, of what you are going to type next. It has no idea of what you are actually thinking about, it is simply making a statistical prediction.
Obsoletes the old idea that "a computer can't be more intelligent than the human who programmed it". Just like a child learning, can eventually surpass the teacher.
An AV knows roughly where it is by consulting GPS signals. They then use an HD local map which records all the static features of area being driven. They then use lidar, radar and cameras to monitor the dynamic environment. Digital cameras are not yet quite good enough to provide accurate 3D modelling of the AV's surrounds, but are improving so fast that some (like Elon Musk), believe that lidars will no longer be needed.
Automated traffic system will assign priorities, usually by people "bidding" for faster travel lanes.
Most of us have never been in a situation where had to choose who or what to collide with in an accident. And even then, nobody wants to articulate their choice of options. They will say "I didn't even think. I just ....". But driverless cars cause consternation bc they force us to publicly reveal this calculation. Society will have to agree on a set of ethical codes to guide decision making in an emergency.
Zero Principle. Disruptive technologies have one common factor: they dramatically reduce a production cost to nearly zero. The steam engine dramatically reduced the cost of running industrial machinery. Before that, plants powered either by water (so had to be beside a fast flowing river) or animals (which have to be fed and continually replaced). Steam power triggered a cascade of innovation. It made steel possible , which in turn made railways possible.
Similarly computers reduced computational cost to near zero and that has flowed through to office work, video games and now, AVs.
Four core costs are reduced to nearly zero by AVs:
1) Toward zero harm: road accidents cost US $18 billion p.a. in hospital costs, and $33 billion in lost wages.
2) Toward zero skill: take out cost of delivery truck or taxi driver.
3) Toward zero time. Indirect cost of today's driving - unproductive time replaced by working or leisure while being transported.
4) Toward zero size: today's cars grossly over-engineered bc of safety needs. A delivery vehicle can be sized according to what it needs to carry, so pizza car can be tiny.
AVs will remove some of the advantages of scale - today bigger companies have a transport advantage bc they can consolidate larger quantities and thus reduce transport costs. But if take out driver's cost, AVs can deliver small quantities from distant producers to a wider range of markets. Any business producing small runs of perishable goods needs a cheap and efficient transport system that doesn't eat into the owner's time.
Traditional maxim that buyers get to choose any 2 out of Cost, Quality and Convenience. But the convenience issue comes out of the equation when AVs can deliver to wherever buyer is - if wanted to buy shoes, several pairs can be delivered to your home, you cjhoose which you want and send rest back.
More books on Cars
Books by Title
Books by Author
Books by Topic
Bits of Books To Impress