MangoBench：面向多智能体目标条件离线强化学习的基准
MangoBench: A Benchmark for Multi-Agent Goal-Conditioned Offline Reinforcement Learning
为视觉驱动的具身智能体提供多目标离线训练与评测基准，其47个任务覆盖高维视觉输入下的联合控制与双臂操作，可直接用于检验视觉表征在稀疏奖励下的目标泛化能力。
原文依据 the first fully cooperative multi-goal benchmark for MARL, covering 3 environments, 4 agent types, and 47 tasks
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论离线多智能体强化学习（MARL）对于在代价高昂且不安全的环境中协调多个智能体至关重要，然而现有方法对奖励函数高度敏感，且对新目标的泛化能力较弱，限制了其实际影响。受单智能体离线目标条件强化学习（OGCRL）的启发，我们提出了首个目标条件离线 MARL 框架，将 OGCRL 扩展到多智能体设定，涵盖完全去中心化以及集中训练与去中心化执行（CTDE）两种范式。为系统评估这一设定，我们提出了 MangoBench，这是首个面向 MARL 的全合作多目标基准，涵盖 3 个环境、4 种智能体类型和 47 个任务，旨在评估联合控制运动、同步与异步双臂操作，以及对高维输入的鲁棒性。大量实验表明，我们的基线方法在稀疏奖励下实现了较强的多目标泛化，但没有任何方法能在所有任务上占据优势，这既揭示了目标条件离线 MARL 的内在复杂性，也展现了其尚未探索的潜力。
英文原文摘要 1110 字符
Offline Multi-Agent Reinforcement Learning (MARL) is critical for coordinating multiple agents in costly and unsafe environments, yet existing methods struggle with high sensitivity to reward functions and weak generalization to new goals, limiting its practical impact. Inspired by single-agent Offline Goal-Conditioned RL (OGCRL), we propose the first goal-conditioned offline MARL framework, extending OGCRL to multi-agent settings under both fully decentralized and centralized training with decentralized execution (CTDE) paradigms. To systematically evaluate this setting, we introduce MangoBench, the first fully cooperative multi-goal benchmark for MARL, covering 3 environments, 4 agent types, and 47 tasks, designed to assess joint-control locomotion, synchronous and asynchronous bimanual manipulation, and robustness to high-dimensional inputs. Extensive experiments demonstrate that our baselines achieve strong multi-goal generalization under sparse rewards, yet no method dominates all tasks, revealing both the intrinsic complexity and the unexplored potential of goal-conditioned offline MARL.