Memory as Plans: World-Action Modeling with Memory-Grounded Planning
作者: Sizhe Zhao, Haozhe Xie, Weiyu Zhao, Chenchu Zhang, Huan Wang, Chenyang Wang, Qinglin Liu, Shengping Zhang
分类: cs.RO
发布日期: 2026-09-10
备注: Project Page: https://sizhezhao.github.io/projects/MaP-WAM/
💡 一句话要点
提出MaP-WAM框架以解决复杂非马尔可夫任务的记忆依赖问题
🎯 匹配领域: 支柱一:机器人控制 (Robot Control) 支柱九:具身大模型 (Embodied Foundation Models)
关键词: 非马尔可夫任务 记忆建模 多模态情境 机器人操作 计划执行 结构化注意力 长短期记忆网络
📋 核心要点
- 现有方法在处理复杂的非马尔可夫任务时,往往无法有效利用长期记忆,导致执行效率低下。
- 本文提出MaP-WAM框架,通过将记忆依赖的建模分解为规划和执行两个阶段,提升了任务执行的灵活性和效率。
- MaP-WAM在RMBench上取得了83.3%的成功率,并在真实机器人任务中达到了78.0%的成功率,展示了其优越的性能。
📝 摘要(中文)
主流机器人策略通常采用马尔可夫模型,但许多复杂的现实操作任务本质上是非马尔可夫的,需超越当前观察的长期记忆。现有的记忆机制往往依赖于语言摘要或视觉窗口,可能会丢失细粒度的视觉证据。本文提出了MaP-WAM框架,将记忆依赖的世界-动作建模分解为基于记忆的规划和计划条件执行,利用长期多模态情境作为规划时证据。MaP-WAM将记忆表示为包含语言指令和稀疏视觉上下文的完成段记录,并将其转化为紧凑的计划,包含下一个段级语言计划和相应的视觉指导。实验结果显示,MaP-WAM在RMBench上取得了83.3%的成功率,并在真实机器人任务中达到了78.0%的成功率,同时在任务历史增长时保持了近乎恒定的推理延迟。
🔬 方法详解
问题定义:本文旨在解决复杂非马尔可夫任务中的记忆依赖问题。现有方法常常依赖于语言摘要或视觉窗口,导致细节丢失和执行效率低下。
核心思路:MaP-WAM框架通过将记忆建模分解为基于记忆的规划和计划条件执行,利用长期多模态情境作为证据,避免了对完整历史的重复条件化。
技术框架:该框架包括两个主要模块:记忆基础的规划模块和计划条件的执行模块。记忆被表示为包含语言指令和视觉上下文的段记录,转换为紧凑的计划。
关键创新:MaP-WAM的创新在于将记忆视为完成段记录,并通过结构化注意力实现关键值缓存,从而在规划和执行中保持上下文长度固定。
关键设计:该框架采用了长短期记忆网络(LSTM)和结构化注意力机制,确保在执行过程中能够有效地更新上下文,同时保持推理延迟的恒定。具体的损失函数和参数设置在实验中经过优化。
🖼️ 关键图片
📊 实验亮点
MaP-WAM在RMBench上取得了83.3%的成功率,相较于现有基线有显著提升。同时,在真实机器人任务中也达到了78.0%的成功率,且在任务历史增长时保持了近乎恒定的推理延迟,展示了其优越的性能和实用性。
🎯 应用场景
该研究具有广泛的应用潜力,特别是在复杂的机器人操作、自动化制造和智能家居等领域。通过提升机器人在非马尔可夫任务中的执行能力,MaP-WAM可以显著提高操作效率和灵活性,推动智能机器人技术的发展。
📄 摘要(原文)
Mainstream robotic policies often adopt a Markovian formulation, but many complex real-world manipulation tasks are inherently non-Markovian, requiring long-horizon memory beyond the current observation. Existing memory mechanisms often rely on language summaries, growing visual windows, or their combinations, and may therefore lose fine-grained visual evidence or face a trade-off between history coverage and execution efficiency. We introduce MaP-WAM, a Memory-as-Plans framework that decomposes memory-dependent world-action modeling into memory-grounded planning and plan-conditioned execution, and uses long-term multimodal episodic context as planning-time evidence rather than repeatedly conditioning the executor on the full history. MaP-WAM represents memory as completed segment records containing language instructions and sparse visual context, and converts this episodic memory into compact plans comprising the next segment-level language plan and corresponding visual guidance. A World-Action-Progress (WAP) model executes each plan over an unknown duration by jointly predicting action chunks and corresponding execution progress at inference time, calibrating predicted progress through plan-observation alignment for adaptive segment transitions and closed-loop context updates. MaP-WAM keeps the executor context length fixed, while structured attention further enables key-value caching in both planning and execution. MaP-WAM achieves state-of-the-art performance on RMBench with an 83.3% success rate and attains 78.0% success on real-robot tasks, while maintaining approximately constant executor inference latency as task history grows.