Signal communication
Only the seer knows which of four pads is the target, and it's locked in a booth. The only way to tell the doer is an 8-symbol channel.
Guessing scores 25%
The short version: neither agent is told to cooperate or talk. Both run one shared policy network and get a single team reward. Cooperation and communication show up only because the rooms make them necessary.
No pretrained models and no predefined symbol meanings. A 3D world built from rooms, simulated 8,192 times in parallel on one GPU.
Only the seer knows which of four pads is the target, and it's locked in a booth. The only way to tell the doer is an 8-symbol channel.
Guessing scores 25%
The door stays open only while someone stands on a plate 7+ units away. One agent holds it for the other, then they swap using a plate just past the door.
Best solo strategy: 0%
The door stays open only while its plate is pressed, and there's no return plate. The way out is to push the cube onto it.
Without the cube: 0.4%
Every room is checked with scripted agents before any training, so the intended skill is the only way through. The numbers above are those scripted checks.
rollout_len = 64 steps. It does this in every one of the parallel worlds at once (--num-worlds 8192 on one GPU, simulated by Madrona on the GPU).update=N in the log counts these rollout→update cycles.Signal took about 270 cycles: 142M world steps, about 18 minutes on one GPU.
Signal success on levels never seen in training, range across 3 training seeds. The last bar pairs a seer and a doer from different runs.
Training success on Signal against world steps, three seeds each.
Tool use is learned the same way: in Cubeplate the trained pair succeeds 98.3% of the time, and every success pushes the cube onto the plate.
Watch with one agent idleHold success on levels never seen in training, 4,096–8,025 episodes per bar.
Why: in every training level, Hold's return plate landed in list slot 0 or 1. In hold+hold it lands in slot 2, and the MLP stops recognizing it.
Compare on hold+holdSuccess on hold+hold, a combination never seen in training. One seed per architecture, about 4,000 episodes per bar.
world steps per second, simulator alone
world steps per second, full training loop
Where one rollout→update cycle's time goes: flat MLP actor, 8,192 worlds, one RTX PRO 6000.
Mute, shuffle or randomize the channel and the team drops to chance.
Idle either agent and Hold becomes impossible. Muting changes nothing.
Unseen hold+hold: a flat MLP against an actor that sees objects as a set.