Motor vehicle deaths are currently running about 33,000 per year in the US. While there are many ways of looking at the numbers, perhaps the most remarkable is we find the carnage acceptable. It is possible to simply reduce the death rate without new technology, but car performance would become unacceptable for most people.1
A big hope is the autonomous vehicle. The National Transportation Safety Agency (NHTSA) has outlined five levels of technology in a 2013 research plan. A rough summary is:
0 current unassisted driving c. 1990
1 Advanced Driver Assistance Systems (ADAS) like anti-lock brakes, electronic stability control, lane-keeping, etc. Common in current cars
2 Integrated functionality like a combination of lane-keeping and adaptive cruise control. Available in some luxury cars now. Human monitoring and the ability to quickly intervene is necessary.
3 Self-driving in some situations like highway driving. Human monitoring and intervention is essential.
4 Full self-driving where unmanned operation is possible.
These definitions are messy, but most people agree upon them as a starting point.2
There is a target, but how and when do we get there? What are the society and technology issues? I'll punt on much of this (let's talk if you're interested), and just focus on a few safety issues that might provide a bit of context.
Level 3 cars will happen in the near future. The capability will be an optional feature on many expensive cars although there will be limited functionality. Highway travel on dry roads is a likely starting space. Although speeds are higher than town driving, the problem space is much smaller most of the time. Cars tend to stay in the same lane and relative speed differences for cars in the same lane tend to be small.
Highway driving is often boring, which is ideal for a computer. The problem is the brief moments of excitement where human intervention will be necessary. Consider the case where the semi you're following brakes heavily and your car's computer decides it can't break in time. A solution might be to veer to one side and pass the truck. This requires knowing road friction, the car's performance, how to introduce torque and information on the new space you will be traveling through. Given sophistication and the appropriate information it could find a safe path for you while not endangering others, but early on it will throw up its hands and wait for you to sort matters out.
In the situation above it would be useful if all of the vehicles in the area would be communicating and creating a joint plan to follow. Vehicle to vehicle communication and a potentially complex executable plan are required. Not only will standards need to be hammered out, but many such maneuvers would require a large percentage of vehicles with at least level 3 functionality. For a long time we'll be at early level 3 where the driver must quickly take control. As the information and computation environment becomes richer we'll move to an enhanced level 3 -- perhaps we'll talk in terms of 3.1 or 3.4 capabilities. If you think Android system fragmentation is an issue, just wait...
Extra safety can be built in by having the system determine that a crash is inevitable and preparing the crash system appropriately. It may make sense to limit speeds and speed variation to give systems better response time to allow the driver to spool up to full capability if a control hand-off is required. We may see very strict 60mph or slower speed limits with strong penalties to encourage all vehicles to blend into the flow. Such tactics may be even more important with early level 4 cars and in situations where weather is involved.
Security is critical and mostly missing in current prototypes. Hacking systems could be a major attraction for a terrorist so this part has to be nearly perfect. It could be the most difficult hurdle to full level 4 autonomy.
City and suburban traffic is more challenging. Extremely good maps and high situational awareness is required. Low speeds would be a big help. European traffic calmed speeds - about 20mph in many areas would be a big help. Requiring conventional vehicles and bike riders to use GPS enabled radios - perhaps their smartphones - to share the road with may become a requirement.
Suburban self-driving is interesting for applications beyond carrying people around. Several companies, including Amazon, are thinking about autonomous ground drones. Lightweight electric delivery vehicles. With appropriate planning 20 mph is just fine for such applications. Delivery capacity is much greater than an airborne drone, costs are potentially lower and public acceptance may be greater. Such vehicles also would create rich photographic datasets that would make autonomous travel easier for manned vehicles and, as such, they may precede city and suburban cars.
The level 3 to 4 transition may be difficult - so difficult that some researchers suggest skipping level 3 entirely. Aviation experience dealing with switching from autopilots to full manual control has proven very difficult. Current recommendations is that someone should be on the steering wheel and brakes as the vehicles begins to get in bad situation. The system may be handling driving and giving a sensation of control and let the human take over as it the computer runs out of options.
Turning your seat around to chat in an autonomous car was predicted by 1960 in 1939 and 2000 in 1964 at respective World's Fairs in New York City. It turns out the relative position of the seat, occupant, and safety equipment is a difficult engineering optimization and making even slight changes can greatly lower occupant safety (eg. don't use an airbag in most cars if you're shorter than 5'0"). Making safety work is a huge challenge - perhaps level 4 cars in a world where most cars are level 4 will crash so little that interior safety is no longer a concern, but that is a long way from where we are now.
While it is very popular to talk about autonomous Uber cars, getting to robust level 4 is a long way off - 15 years before we get to anything like 5% penetration srikes me as extremely optimistic, although I wouldn't be surprised to see 20 or 30% level 3 by then. I am assuming 'simple' issues like laws and insurance can be solved by then and that the difficult tasks like security are well understood. Level 4 offers fascinating society and technology issues to think about, but focus on them is not understanding some larger problems. In the meantime there is a long path to follow and much to be learned.
1 The kinetic energy of a car goes as the square of its velocity. Limiting top speed could dramatically improve crash survivability as well provide more reaction time for the driver. Some researchers suggest current safety systems in current cars would be safe for most sub 35 mph collisions.
2 The plan is here (pdf) There are some criticisms as it suggests a technology hierarchy where each level is a superset of the previous. Talking about level 3 and level 4 in any detail is a very complex task. It is possible to imagine level 4 operation in some conditions that are much easier than solid level 3 in others. Ultimately levels are probably the wrong way to look at the problem. They are only meaningful in terms of the type of road and driving task. Driving tasks involve hundreds of decisions rather than something simple like 'drive the car'. Like any complex technology it is messy with a lot of qualifiers. Hopefully this will be sorted out more meaningfully as technologies develop. my own suggestion:
° hands free
° driver not watching (eyes free)
° unmanned automatic with a network connection
° fully autonomous without network connection
° full capability/drive anywhere expertly without network connection
Pasta, Tomatos and Asparagus
° olive oil
° 2 cloves of garlic finely sliced
° small bunch of basil finely chopped
° one 28 oz high quality Italian plum tomatoes (until we get the real thing from the market)
° 1 bunch asparagus
° 1 pound long flat pasta like tagliatelle
° salt and freshly ground pepper
° Parmesan cheese for serving - leave off if vegan
° Heat a sauce pan with a glug of olive oil to medium heat. Sauté garlic 'til golden. Add some of the basil and tomatoes, season with a bit of salt and pepper and bring to a boil. Cut the heat to a simmer and go about five minutes
° Snap off the woody ends of the asparagus and chop the rest. Add it to the sauce and simmer for a short time. Remove from the heat. Finish seasoning and add the rest of the basil.
° cook the pasta and reserve a bit of the water. Now toss it with the sauce and just a bit of the water. Drizzle on an appropriate amount of olive oil and garnish with a lot of Parmesan if you like.