Drones and self-driving basicsMiddleLesson 2878 min read

Self-driving cars: what is real today

Driver assistance is common, and driverless taxis run in a few places under limits. Learn the levels of automation, why rare situations are so hard, and why Indian roads are a special challenge.

✦ Ask the tutor 🧪 Lab · 25 min 📝 Lesson test

Speed

Lesson in motion

In 60 seconds

Self-driving cars: what is real today

Driver assistance is common, and driverless taxis run in a few places under limits. Learn the levels of automation, why rare situations are so hard, and why Indian roads are a…

1/4

In simple words

Learning to cycle in an empty park is easy. Cycling in a busy market is much harder: a cow walks out, a child runs after a ball, an auto stops suddenly, a shopkeeper waves you through. A car that drives itself has to handle all of this, and also the strange things that almost never happen. The rare things are the hardest part.

You may hear that self-driving cars are here, or that they will never work. Both are too simple. Some parts of driving are already automated well. Full self-driving everywhere, in every condition, is not available today.

Levels of driving automation

Engineers commonly describe driving automation with a scale of six levels, from 0 to 5. The exact definitions come from an engineering standards body; this is the general idea:

Level	Who drives	In plain words
0	Human	The car may warn you, but you do all the driving
1	Human, with one kind of help	The car helps with either speed or steering, for example keeping a distance from the car ahead
2	Human, with more help	The car can steer and control speed together in some conditions, but you must watch the road all the time and be ready at once
3	The car, in limited conditions	The car drives in specific conditions; you need not watch all the time, but must take over when the car asks
4	The car, in a limited area	No human needs to take over inside a defined area and conditions; outside them, it will not drive itself
5	The car, everywhere	Drives anywhere a human could, in any condition. This does not exist today

What is real today

Driver assistance (levels 1 and 2) is common in many new cars, including in India: automatic emergency braking, lane keeping, adaptive cruise control. The human is still the driver.
Driverless ride services (level 4) operate in a small number of cities in some countries, inside mapped areas and conditions chosen by the operator. They are closely monitored and are still expanding carefully.
Controlled places such as mines, ports, warehouses and campuses are where autonomy works best, because the space, speed and traffic are limited.
Level 5, a car that can drive anywhere in any condition, is not available anywhere.

Watch out

Names used in advertising, like "autopilot" or "self-driving", do not tell you the level. Read the car's manual. If it says you must keep your hands on the wheel and eyes on the road, you are the driver, whatever the feature is called.

Why rare situations are so hard

Most driving is routine: follow the lane, keep a distance, stop at red. A system can learn that from lots of data. The trouble is the long tail: thousands of different rare situations, each of which happens seldom, but together happen often. A tyre rolling across the road. A policeman waving traffic through a red light. A mattress falling off a truck. A procession blocking a street.

Rare events are rare in training data too, so the system has little to learn from. Simulation helps, because engineers can create many rare situations on purpose, but a simulator only contains what someone thought to put in it. This is the same gap you met in Simulation and sim-to-real.

Indian roads are a special test

Situation	Why it is hard for a machine
Mixed traffic	Cars, buses, bikes, autos, cycles, carts, pedestrians and animals share the same road, each moving differently
Faded or missing lane markings	Lane-keeping systems often depend on clear markings
Informal signals	Hand waves, horns, flashing headlights and eye contact carry meaning that is hard to read
Potholes, speed breakers, waterlogging	Hard to see, especially at night or in the monsoon, and not always on any map
Wrong-side driving, sudden U-turns	Other road users may not follow the rules the system expects
Changing roads	Construction and diversions change the road faster than maps are updated

None of this makes autonomy impossible in India. It means that systems built and tested elsewhere cannot be assumed to work here, and need testing on Indian roads, in Indian weather, with Indian traffic.

Human responsibility

With driver assistance, the person in the driver's seat is responsible for driving. The system can switch off without much warning, for example when it cannot see lane markings, and the human must be ready at every moment. A big risk is over-trust: the system works well for many kilometres, the driver relaxes and looks at a phone, and then a rare situation arrives. Laws on who is responsible for more automated vehicles are still developing in many countries; check the current rules where you live.

Over-trust

"The car drives itself, I can check my phone"
Believing the feature name
Using assistance in heavy rain or on unmarked roads

Healthy use

Hands ready, eyes on the road, all the time
Reading the manual for what the feature cannot do
Switching assistance off where it struggles

Try this. On your next trip by bus or auto, list five things that happened that a self-driving system would find hard. For each, write which sensor or which part of the system (sensing, localisation, prediction, planning) would be challenged.

Watch and read more

Lab

A long-tail list for one real Indian road, with the system part each case would stress and a safe fallback.

~25 min

The problem

No coding needed. Choose one road you know well, for example the road outside your school, college or market.

List at least ten situations on that road that a self-driving system would find hard. Include at least three rare ones you have seen only once or twice.
For each, write which part is challenged: sensing, localisation, prediction of others, or planning.
For each, write a safe fallback: slow down, stop, hand over to a human, or avoid that road or time.
Finish with three sentences: which level of automation, if any, would you trust on this road today, and why?

You are done when

At least ten situations, specific to one real road, including three rare ones
Each situation is linked to the right system part (sensing, localisation, prediction, planning)
Fallbacks are safe and realistic, and never assume the system is always right
The final answer is honest and gives reasons, not a guess about future products

Hard questions

Try to answer before you reveal. If you can answer these, you understood the lesson.

Q1Why can't engineers just drive enough kilometres to see every rare situation?Reveal

Because the number of possible rare situations is enormous, and each happens so seldom that even a lot of driving may see some only once or never. Simulation lets engineers create rare cases on purpose, but it only contains what someone thought to include. So teams combine real driving, simulation, careful limits on where the system runs, and fallbacks for when it is unsure.

Q2Level 3 lets the driver look away but asks them to take over when needed. What makes this hand-over hard?Reveal

A person who was not watching needs time to understand what is happening before acting well. If the system asks for help only when it is already in trouble, there may not be enough time. That is why the design of the hand-over, how early it warns and what it does if the human does not respond, matters as much as the driving itself.

Please sign in to continue.

Questions people ask

Are self-driving cars safer than human drivers?

This is argued about a lot, and the answer depends on the system, the place, the conditions and how you count. Honest comparisons are hard, because driverless systems run only in chosen areas and conditions. Be careful with any simple claim either way, and look for independent studies with clear methods.

When will we have level 5 cars?

Nobody knows. Predictions have often been too optimistic in the past. Progress is more likely to come step by step: more areas, more weather conditions, more kinds of roads.

Can I legally use a driverless car in India today?

Driver-assistance features are sold in India, and you remain the driver. For anything more automated, check the current rules from the Ministry of Road Transport and Highways and your state; laws in this area are still developing.

Do self-driving systems use AI like chatbots?

They use a lot of machine learning, especially for seeing and predicting what others will do, and some newer approaches learn larger parts of driving end to end. But they also rely on careful engineering: sensor fusion, maps, planning, safety checks and fallback plans. A wrong answer from a chatbot is text; a wrong decision on the road can hurt people, so the bar is much higher.

Lesson test

5 questions. Get 3 right (60%) to pass and complete this lesson.