Sensor Fusion 101
why a car trusts three liars
A car needs to know where it is. Every sensor it has is wrong in its own way: GPS is unbiased but noisy (and dies in tunnels), wheel odometry is smooth but drifts without bound, lidar against a known map is precise but conditional (it needs mapped landmarks actually in view — in range and not hidden). Sensor fusion is the art of combining wrongs into one estimate that is better than any single source. This lab makes that visible: the gold car is the truth, the purple pad is what odometry believes, the blue pad is the fused estimate, and the glowing disc under it is how unsure the filter admits to being.
What you can try
Each preset is a prepared situation, and the lab says what is worth noticing in it:
- open sky — all three sensors healthy — watch how differently the filter weighs them
- tunnel — the walls block every satellite and hide every landmark: the filter coasts
- lidar out — the precise sensor is gone — now the GPS noise reaches the estimate
Running it
Labs run in the Dojo from a Clayground checkout:
cmake -B build && cmake --build build
./build/bin/claydojo --sbx labs/sensor-fusion-101/Sandbox.qml
Press ? inside any lab for its key map, and T where a guided tour is offered.