Self driving cars

 

Self driving cars

There has been a lot of press talk about this predicting that many thousands of truck drivers will become redundant. Also recently UK government has announced plans to allow limited self driving on UK motorways.

In describing self driving the US Auto standards association recognises 5 levels ranging from Level 1 automatic lane keeping through to level 5 full self driving. No vehicle is presently authorised for routine use at level 5 although some development use is permitted.

There has been use of driver aids which permit a car driver to drive more safely. For example lane departure warnings. My ten year old car has adaptive cruise control which is tied to a crude heads up display warning of proximity to cars ahead and giving an indication of closing speed. The only automatic function is emergency braking to mitigate possible collision. This is all taking data from a forward facing radar.

The key automatic functions now found on most cars are anti lock braking ( usually denoted ABS after the German )and dynamic stability control. Anti lock braking is fairly self explanatory. It functions by releasing the brake pressure at the point of wheel skidding and reapplying it again. This cycle is repeated many times very quickly. Anti lock braking does slightly increase braking distance but the compensation is retention of vehicle control as a skidding car can much increase stopping distance.

Dynamic stability is the automatic selective application of brakes to help ensure the vehicle goes where the driver intends it to go without skidding. For example to limit the tendency of the car to understeer ( ie go straight on .unless more steering lock is applied ) the inner rear wheel is braked . Similarly other wheels are individually braked according to circumstance. This does not improve the cars cornering capability but ensures that driver control is retained. Since loss of control due to skidding is the fundamental cause of one third of all accidents this is a massive safety boost. Use of this technology is mandatory for all new cars. The technology comes under many names such as electronic stability control ( ESC )

Self driving vehicles were originally sponsored by DARPA ( The US Defence Advanced Research Projects Agency ) .  The trials in the Mojave desert eventually showed the technology was possible and commercial development began about 2010. The acknowledged leader in autonomous vehicles is Waymo, a subsidiary of Alphabet the parent of Google. Waymo has collaboration agreements with several major car companies including Nissan and Volvo

Waymo have tried building dedicated vehicles as Robo-taxis but seem to have settled for modifying some electric vehicles such as Chrysler minivans. Essentially these vehicles are equipped with cameras and LIDAR ( a form of radar using laser light ). LIDAR is claimed to be able to detect objects up to 300m distant. In addition short range radar detects closer objects. Waymo is operating at Level 4 which is autonomous on public roads with other users but in conditions of good climate, particularly visibility.

In addition to various test areas in the US Waymo operates a test commercial service in Phoenix, Arizona. Phoenix road system has been fully mapped into the cars computer. Residents can call up the car and be driven where they want in the city environs. There is  usually a safety driver although not always.

In addition to Waymo there is CRUISE, a General Motors autonomous driving company and another associated with Uber

The most different is Tesla the electric car company of which Elon Musk is chairman. They are taking a somewhat different approach. Tesla do not use LIDAR ( which is expensive ) or even radar but rely on a suite of cameras around the vehicle. For some time Tesla cars have been available with Autopilot as an option. Autopilot allows for automatic lane changes . This has been developed to a prototype full self driving system. This is currently in beta test which means it is available to a few Tesla owners for test and evaluation. The objective is an autonomous system which doesn’t require road pre mapping.

All these systems rely on artificial intelligence( AI ) as part of the computing. This simply means training the computer system to think like a human. Think of this like a network. It learns by allowing reinforcement of the paths leading to a correct solution. In its training phase the computer is presented with a picture. Lets say we are teaching the computer to differentiate between a cat and a dog. If the computer gives the right answer the network is reinforced, if wrong the pathway is  weakened. The decision is made by a human  whether the computer is right or wrong. By looking at many thousands of images the computer learns by reinforcement what is right and what is wrong. Correct training leads to the computer knowing the characteristics of a dog from a cat and makes the correct choice. On a massively larger scale this is what the driving computer does.

In Tesla practice for full self driving there are two computers side by side which have to agree on the decision. Training the computer is essential which means that many thousands of images must be presented and interpreted. It so happens that Tesla cars have always been fitted with cameras, initially to aid the driver, but this means that Tesla has millions of hours of real driving images for training. This is sometimes seen as a key benefit for Tesla relative to its competitors.. The importance is the occurrence of unusual situations. For example a tow truck might tow a broken down truck with its rear wheels suspended so the recovered vehicle is facing backwards. It is essential that this is not confused with a vehicle being driven  the wrong way on a divided carriageway.  Many other situations occur rarely in driving but it is essential that the AI is trained for all of them.

It seems clear that at present Tesla is ahead of the industry in testing self driving. Their objective is a system which will work on any road without any prior mapping and also one which is affordable. The existing charge is $10000 but this is for a more limited level 3 system. It looks as if the rate determining step will be regulation not technology.