RECORDED RUN / 24 S LOOP

A robot's run, rebuilt by drones.

A recorded run from the Zonta Labs research simulator. A small robot car drives a track in a lab while simulated drones watch it and rebuild the room and the car in 3D.

The run plays back in your browser from a recording. It is research in simulation, not a shipped product.

RECORDED SIMULATION / DRONES MAP A LAB TRACK IN 3D

Controls

DRAG
Orbit the camera
RIGHT-DRAG
Pan
W A S D
Fly forward, back and sideways
Q / E
Fly down and up
SHIFT
Fly faster

Click the view first so it takes the keys. On a touch screen the run plays without controls.

What you are seeing

  1. The drones

    52 simulated drones. Eight fly with the car and aim their cameras at a marker on its roof. The other 44 sweep the room.

  2. The map

    The room and the car appear only where a drone camera has seen them. Anything no camera has reached stays blank, so the map fills in as the run goes on.

  3. The car

    The car has no sensors of its own. A filter fuses the drones' noisy measurements into one estimate of where it is, and its driving policy steers from that estimate. The car is drawn at the estimate, never at its true position.

  4. Where it fits

    This is the first stage of the loop we are building: watch a run and rebuild it in 3D. Handing that reconstruction to AI agents, so they can explain a failure and improve the next run, is the part still in development.

  5. Limits

    The simulation is simple on purpose. The car has no tire slip, the camera measurements come from a hand-written sensing model rather than rendered images, and none of it has been tested on hardware yet.

Research recordings