Autonomous driving software is the software that recognises what a car's sensors detect, plans the path and controls the steering and brakes, so that the car drives with no human driving it. It is also called the self-driving stack, because it is built in layers.
The software does three jobs. First, perception: it uses the readings of the cameras, radar and lidar to recognise a car, a scooter or a person. Second, planning: it decides where the car should go in the next few seconds. Third, control: it turns that plan into steering and braking commands.
This product is sold mainly abroad. Companies that build full self-driving software sell it mostly in the United States and China, where driverless taxis are allowed in some cities. There is no driverless taxi service in India. Indian law, the Motor Vehicles Act, assumes that a human is driving, so a car with no driver is not yet allowed on Indian roads. Indian engineers do work on parts of it, such as testing and maps.
Consider a self-driving taxi in a city in the United States, such as Phoenix. It comes to a junction at 30 miles per hour (about 48 km/h). Perception finds a pedestrian near the kerb and a bus stopping ahead. Planning decides to slow down and wait for the pedestrian. Control reduces the speed to a stop, smoothly. This repeats many times every second.
Software of this kind used to follow rules written by hand. Now many companies use large AI models instead, trained on millions of hours of driving. Some use a single model that learns to go from what the cameras see straight to steering and braking. Such models are related to foundation models. Indian roads, with scooters, cattle and unmarked lanes, are among the hardest in the world for this approach.
It is different from driver assistance. Driver assistance helps a human who is driving and responsible. Autonomous driving software is responsible for the driving itself, so it must also decide what to do when something fails.
The software also depends on an HD map that shows each lane. It is tested over millions of virtual kilometres in simulation before it drives a real road. Driverless taxis, called robotaxis, use it today. Drones use similar methods, described under drone autonomy software. The levels of autonomy describe how much of the driving it does. Because it can cause injuries, it needs safety certification.