EmbodiedMemory-Bench：面向长时程具身任务的记忆基准
原标题：Paper page - EmbodiedMemory-Bench: Benchmarking Embodied Memory for Long-Horizon Embodied Tasks
AI 摘要
研究者提出 EmbodiedMemory-Bench（EMem-Bench），一个包含 2,554 个交互 episode、覆盖四类任务家族的基准，用于评估长时程具身任务中的记忆能力。论文指出当前智能体存在四类记忆缺陷：细粒度视觉记忆弱、动态世界状态跟踪不可靠、无法记录交互结果揭示的世界状态、以及经验泛化能力有限。作者同时提出外部记忆系统 Embodied-Memorizer（EMem）及基于 8B 策略的 EMem-8B，实验显示现有开源与专有多模态大模型在该基准上表现薄弱且不均衡，而 EMem 在匹配骨干下取得最佳整体表现。
正文节选
EmbodiedMemory-Bench: Benchmarking Embodied Memory for Long-Horizon Embodied Tasks Abstract Long-horizon embodied interaction requires agents to retain and continually update information about the environment as they observe, act, and encounter change. Yet current agents struggle to maintain such memory reliably. Our analysis traces this limitation to four key deficiencies: weak fine-grained visual memory, unreliable dynamic world-state tracking, failing to record world state revealed by interac