OBSTACLE RACE CHAOS RUN
Start with the basics, then take on tougher courses.
LOCAL · 0/6 · NO REWARDS
YOUR ROBOTGUEST DUCK
PLACE1 / 6
L1 · ROOKIECP 0 / 1
RUN TIME00:00.00
Factory Run
Waiting for runners.
Shared rooms · up to 6 players
3
LOADING THE COURSE…
WASD / ↑↓←→ MOVESPACE JUMPSHIFT DIVER RETRY
ROBOTS League 3D multiplayer duck soccer
GREEN0
FIELD IDLE01:30
0YELLOW
ROBOTS / 3D ARENA
GUEST DUCK / PRACTICE UNIT
Loading duck models…
WASD MOVE
SPACE KICK
SHIFT DASH
LOCAL SIM
MOVE LAB / TACTICAL PROTOTYPE
TEACH A MOVE.
PROVE IT WORKS.
Tune a bounded soccer behavior, search for improvements locally, then publish the policy and its test evidence. Prototype points come from held-out simulation results—not playing time.
EVIDENCE PIPELINEPOINTS ≠ MONEY
- BUILDValidated move schema. No uploaded scripts or arbitrary code.
- TRAINLocal bounded policy search produces a versioned artifact and replay trace. Cloud RL is not connected yet.
- CHALLENGEServer-only seeds vary ball friction, control delay, sensor noise and opponents.
- PUBLISHOnly reproducible improvement creates a public manifest and contribution points.
JOBNOT STARTED
HELD-OUT SCORE—
POINTS—
MANIFEST—
WHAT “USEFUL” MEANS HERE: a move must improve the lab’s held-out tactical benchmark and survive domain randomization. This separate prototype does not yet train the 3D match ducks or prove transfer to a physical robot. That requires a shared simulator, calibrated physics and real-robot tests.