Rhythm Sync

Agents in separate rooms must press paired switches inside a shared timing window that is cued on screen. Pressing alone or out of time fails the stage. Some stages give the cue to only one agent, so the other has to infer the timing.

import comrad

env = comrad.parallel_env("rhythm_sync")

Gymnasium id	comrad/rhythm_sync-v0
Agents	2 by default (supports 2)
Capability tested	Temporal coordination, implicit synchronisation
Episode limit	1750 steps
Observation	RGB image (72, 128, 3) per agent; measurement vector available with additional_input=True
Action space	Tuple(Discrete(3), Discrete(3), Discrete(3), Discrete(2))
Default shared_reward	1.0: rewards are fully shared across the team
Score (info["true_objective"])	Stages completed

Actions

Each agent sends one choice per head; index 0 is always "do nothing".

1. move: none / forward / backward
2. strafe: none / right / left
3. turn: none / left / right
4. use: none / use (press switch)

Reward

+1 per newly completed stage and -1 per failed stage, split across agents.

The score in info["true_objective"] is what the benchmark reports. The shaped reward is only a training signal.

Variants

• rhythm_sync_dense: same map with dense reward shaping.