Sensor Lab
How a self-driving car sees the road
A self-driving car has no eyes. It has sensors, and each one is good at some things and bad at others. Turn them on one at a time, then make the weather worse and watch what happens.
Step 1 of 6
Meet the test car
Hi! This little car has three ways to see the road. Ahead of it are a person, a cyclist, a broken-down truck and a sign. Let's find out who sees what.
LiDAROff
Laser pulses that measure distance in 3D
Sharp 3D dots out to about 110 m.
Reach69%Detail85%RadarOff
Radio waves that measure distance and speed
Long reach and speed, but just blobs.
Reach100%Detail20%CameraOff
Pictures that an AI model reads
Sees colour, lane lines and reads the sign.
Reach75%Detail100%
What the car knows
Road closed sign30 m
- LiDAR: off
- Radar: off
- Camera: off
Result: Not seen in time
Pedestrian42 m
- LiDAR: off
- Radar: off
- Camera: off
Result: Not seen in time
Cyclist55 m
- LiDAR: off
- Radar: off
- Camera: off
Result: Not seen in time
Stalled truck75 m
- LiDAR: off
- Radar: off
- Camera: off
Result: Not seen in time
Good to know
Why real cars carry all three
Each sensor fails differently
Cameras need light and clear air. LiDAR copes with darkness but not with thick rain or fog. Radar copes with weather but can't tell a person from a post. Put together, their weak spots rarely line up.
Spotting a thing and naming it are separate jobs
LiDAR and radar report that something is there. Working out that it's a child, a bin bag or a sign that says the road is closed is a separate job, usually done by AI models trained on huge numbers of labelled examples.
Stopping takes room
At 72 km/h a car covers 20 metres every second. A computer reacts quickly, but the brakes still need room. In this lab the car needs about 39 metres to stop, so a sensor that sees 30 metres ahead is not enough.
Your turn
Night Drive
Drive home through sun, rain, fog and darkness with only one sensor powered at a time. Pick the wrong one and you won't see the danger until it's too late.
Play Night Drive →