Platform Chain

Tethered players climb a long route of platforms over a hazard. Moving too far apart breaks the tether, so both must keep close while jumping. Checkpoints only advance in order.

import comrad

env = comrad.parallel_env("platform_chain")

Gymnasium id	comrad/platform_chain-v0
Agents	2 by default (supports 2)
Capability tested	Long-horizon joint movement under a tether
Episode limit	5250 steps
Observation	RGB image (72, 128, 3) per agent; measurement vector not available
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"])	Joint checkpoint progress (max 47 on the default map)

Actions

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

1. turn: none / left / right
2. move: none / forward / backward
3. strafe: none / left / right
4. jump: none / jump

Reward

+1 per level of best joint progress; penalties for chain breaks, dragging the chain and death.

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

Variants

• platform_chain_easy: shorter, easier route.
• comrad/scenarios/batch_platform_chain_curriculum/: curriculum pool used in the paper (see Reproducing).