Probing Dec-POMDP Reasoning in Cooperative MARL

TL;DR

A diagnostic suite combining statistically grounded performance comparisons and information-theoretic probes to audit the behavioural complexity of baseline policies across 37 scenarios spanning MPE, SMAX, Overcooked, Hanabi, and MaBrax reveals that success on these benchmarks rarely requires genuine Dec-POMDP reasoning.

Abstract

Cooperative multi-agent reinforcement learning (MARL) is typically framed as a decentralised partially observable Markov decision process (Dec-POMDP), a setting whose hardness stems from two key challenges: partial observability and decentralised coordination. Genuinely solving such tasks requires Dec-POMDP reasoning, where agents use history to infer hidden states and coordinate based on local information. Yet it remains unclear whether popular benchmarks actually demand this reasoning or permit success via simpler strategies. We introduce a diagnostic suite combining statistically grounded performance comparisons and information-theoretic probes to audit the behavioural complexity of baseline policies (IPPO and MAPPO) across 37 scenarios spanning MPE, SMAX, Overcooked, Hanabi, and MaBrax. Our diagnostics reveal that success on these benchmarks rarely requires genuine Dec-POMDP reasoning. Reactive policies match the performance of memory-based agents in over half the scenarios, and emergent coordination frequently relies on brittle, synchronous action coupling rather than robust temporal influence. These findings suggest that some widely used benchmarks may not adequately test core Dec-POMDP assumptions under current training paradigms, potentially leading to over-optimistic assessments of progress. We release our diagnostic tooling to support more rigorous environment design and evaluation in cooperative MARL.

Figures (14)

• Figure 1: Summary of information-theoretic diagnostics. Colours denote source (blue), target (green), and conditioning (amber, dashed) variables. All quantities are expectations under the converged joint policy $p^{\boldsymbol{\pi}}$; $O_t^i$ is agent $i$'s observation, $A_t^i$ its action, $H_t^i$ the local history (RNN hidden state or finite window), $\tau^{i}_{t-1}$ agent $i$'s action--observation history up to $t{-}1$, and $T$ the episode horizon.
• Figure 2: Simple Reference
• Figure 3: Speaker Listener
• Figure 4: Simple Spread
• Figure 5: $\Delta_{\text{Mem}}$; bold: $p{<}0.05$.

Theorems & Definitions (2)

• definition 1: Relevant Partial Observability
• definition 2: (Markov) Equilibria