(Originally published on LinkedIn in June 2024)
You already know the answer to this. Just think about how humans drive – Step 1: get drunk, step 2: bet your friend that you can beat him/her in a race to the shopping center, step 3: …
Just kidding. This is how we humans actually drive – 1) Perception: we perceive the environment (the road, cars and pedestrians on the streets, traffic light etc) through our senses – eyes, ears, and our bodies (for example – getting a “feel” of a slippery road by the “movement” sensor inside us). 2) Decision: based on the perceived data about the environment, our brain decides to steer, accelerate, hit the brakes, use the indicator or even give a finger. 3) Action: based on the decision of the brain, our limbs then make the needed movement – shifting the gear, pressing the brake/gas pedal, rotating of the steering wheel, pushing the arm out of the window and holding up the middle finger.
And these are exact three things that an autonomous vehicle does too – perception through sensors (cameras, radars, ultrasonic sensors and such), decision through brain (the computer within vehicle’s electronics), and action through “actuators” (the electronic and mechanical devices in the car to execute the steering and (ac-)decelerating. “That sounds simple!” you might say. “Yes, but the devil lies in the details!” says the engineer.
There are thousands of researchers working on just one aspect of one type of sensor, for example – object classification by a camera (i.e. recognizing if a group of pixels in a camera image is a pedestrian, another vehicle, a dog, a lamp, or something else, an alien?). And there are thousands more working on just radar, or the decision tree of the automation algorithm, or the simulation model of a traffic environment, or a smooth control of the car, or … you get the point.