Definition
In transport, AI powers route planning, traffic prediction and driver assistance, up to self-driving vehicles. In manufacturing, it powers visual quality inspection and predictive maintenance. SAE defines six levels of driving automation, from 0 (none) to 5 (full).
In simple words
Maps guess your arrival time from traffic. Factory cameras spot scratches faster than the human eye. Sensors notice that a machine sounds wrong and warn before it breaks.
Why it matters
These are large, practical uses of AI. Self-driving also shows how the last few percent of reliability can be the hardest part of all.
Key terms
- Autonomous vehicle
- A vehicle that can drive itself, at least in some conditions, without a person steering.
- Driver assistance
- Features that help a human driver, such as automatic braking, while the driver stays responsible.
- Lidar
- A sensor that measures distance by sending out laser light and timing its return.
- Radar
- A sensor that uses radio waves to find how far away objects are and how fast they move.
- SAE levels
- A scale from 0 to 5 that describes how much of the driving a vehicle does by itself.
- Edge case
- A rare situation on the road that a system has seldom or never seen.
Explained in detail
How a self-driving car works
A self-driving system repeats four steps many times a second. First it perceives: cameras read lane lines, signs and traffic lights; radar measures the distance and speed of other vehicles, even in fog; lidar builds a three-dimensional picture of everything nearby; and a detailed map with satellite positioning tells the car where it is. Second it predicts what each road user is about to do: will that cyclist turn, will that child step off the kerb? Third it plans a safe path and speed. Fourth it controls the steering, accelerator and brakes to follow the plan.
A short history
In 2004 a research agency of the United States government held a race for driverless vehicles across 240 kilometres of desert. No vehicle finished; the best travelled about twelve kilometres. Eighteen months later, in the second race, five vehicles completed the course. That rapid progress convinced many people that self-driving cars were only a few years away. Twenty years on, driverless taxis do carry paying passengers in parts of a few cities in the United States and China, but only within mapped areas and in suitable weather. The last part of the problem has proved far harder than the first.
Why the last part is so hard
Ordinary driving is mostly routine, and AI handles routine well. The difficulty lies in rare events: a mattress lying in the lane, a police officer waving traffic through a red light, a wedding procession, a herd of buffalo. Each such event is rare, but there are endless kinds of them, so a car meets something unusual quite often. A human driver uses common sense. A model has only its training data. Roads in India add further difficulty: lanes are loosely followed, and cars share the road with motorcycles, auto-rickshaws, carts, pedestrians and animals.
What the levels mean for you
Most cars sold today with self-driving-sounding features are at Level 2. The car can steer and control its speed on a marked road, but the human must watch the road all the time and is responsible for everything that happens. The danger is that the system works so well, so much of the time, that the driver stops paying attention. In 2018 a test vehicle operating in self-driving mode struck and killed a pedestrian in the United States; the safety driver was looking away from the road. Investigators pointed to over-trust in automation as a cause.
AI elsewhere in transport
- Safety features in ordinary cars: automatic emergency braking, lane-keeping help and warnings when the driver looks drowsy.
- Traffic signals that adjust their timing to the queues they see.
- Railways and metros: predicting which parts need maintenance before they fail.
- Delivery companies: planning the order of hundreds of stops to save fuel and time.
- Ports and airports: scheduling cranes, gates and runways.
How it works
| SAE level | Who is driving |
|---|---|
| 0 | The person. No automation. |
| 1 | The person, with steering or speed help |
| 2 | The person watches, while the car steers and controls speed |
| 3 | The car, in set conditions, and the person must take over on request |
| 4 | The car, within a limited area, with no person needed |
| 5 | The car, everywhere |
Try it yourself
Handing over the steering wheel
Move up the six levels of driving automation, and watch who is responsible.
Who is driving?
Match each car to its SAE automation level. The table above will help.
- An old car with no driving help at all
- The car only keeps the speed. The person does everything else
- The car steers and controls speed while the person keeps watching
- The car drives in set conditions and may ask the person to take over
- A taxi with no driver that runs only inside one city area
- A car that can drive itself anywhere, with no person needed
Common mistakes
- Trusting driver-assist features as if the car drives itself.
- Assuming models trained abroad handle Indian roads.
Check yourself
Summary
- Routes, traffic, inspection and maintenance are widely used now.
- SAE levels run from 0 to 5.
- Most driver-assist cars still need a watching driver.
Finished reading?