Self Driving Car Robotics

Understanding Self Driving Car Robotics

Each sensor has strengths and limits. Cameras capture colour, lane paint, sign text, and traffic signal states. Their performance can fall in glare, darkness, fog, heavy rain, or when a lens is dirty.

Lidar sends out laser pulses and measures their return time to build a detailed three-dimensional shape of nearby objects. It can estimate distance well, but rain, snow, dust, and reflective surfaces can create unreliable returns. Radar uses radio waves.

It works well through poor weather and directly measures how quickly an object is moving toward or away from the car. Its image of object shape is less detailed.

Ultrasonic sensors are useful at very short range, especially during parking. A vehicle combines these sources because no single sensor provides a complete dependable view.

The computer must turn raw measurements into a model of the road. It groups pixels, laser points, and radar reflections into possible objects. It then tracks each object over several moments.

Tracking helps distinguish a parked vehicle from a cyclist moving into the lane. The system estimates position, speed, direction, and uncertainty. Uncertainty matters because sensor data is never perfect.

A shadow may look like a hole, a plastic bag may resemble an obstacle, or part of a pedestrian may be hidden behind a van. Safe software does not pretend every estimate is exact. It leaves extra space, slows down when it cannot see far ahead, and chooses cautious actions when the evidence is unclear.

Knowing where the car is is a separate robotics task called localisation. Satellite navigation gives a broad position, but it can be inaccurate near tall buildings, tunnels, and trees. Cars improve this estimate by comparing camera or lidar observations with detailed maps and by measuring wheel rotation and vehicle motion.

These motion sensors gradually drift, so the car must correct them using outside landmarks. In a roadwork area, an old map may no longer match the real lane layout.

The car therefore needs to give more weight to current sensor observations than to stored map information. Students can connect this idea to a phone map that briefly places them on the wrong street when its signal is weak.

Planning is more than drawing a line to a destination. The car predicts several likely movements for nearby road users, then checks possible paths against road boundaries, traffic rules, comfort, and collision risk. It needs to leave enough time to brake.

Since stopping distance grows with the square of speed, doubling speed can require about four times the braking distance when deceleration stays the same. Control software then converts the selected path into small steering, braking, and acceleration commands. Sudden commands can make passengers uncomfortable or cause loss of grip on wet roads.

Automation levels describe who is responsible for the driving task in particular conditions, not a simple scale of intelligence. Many systems work only on approved roads or in certain weather. A useful habit when studying this topic is to ask what conditions the system assumes, what it can fail to detect, and how it reaches a safe fallback state.