ScholarPulse 日报 2026-10-06

2026-10-06 学术简报：2 篇。RV-ICL是一种训练免费方法，将演示视频转化为智能体可导航的层次结构而非直接提示，从而提升冻结VLA策略的任务成功率。

今日速览

序号	标题	来源	日期	主题	推荐等级
1	Recursive Video In-Context Learning for Agentic Robot	arXiv	2026-10-05	AI-Agent	高
2	MemPilot: Orchestrating On-Demand Multimodal Memory Curation for LLM Agents	arXiv	2026-10-05	AI-Agent	高

重点论文与技术动态

1. Recursive Video In-Context Learning for Agentic Robot

来源：arXiv
日期：2026-10-05
作者/机构：Wenrui Bao, Xinxin Liu, Bingxin Xu, Yuzhang Shang
主题标签：AI-Agent, arXiv
推荐等级：高
分类：cs.RO, cs.AI, cs.CL, cs.MA

一句话结论

RV-ICL是一种训练免费方法，将演示视频转化为智能体可导航的层次结构而非直接提示，从而提升冻结VLA策略的任务成功率。

核心内容

LLM智能体通过文本记忆仅记录“做了什么”，无法捕捉任务如何完成。
演示视频难以适配上下文：全视频拖慢每轮交互，固定关键帧丢失决定抓取成败的接触细节。
RV-ICL从抓取、释放等子事件构建层次，粒度从整体关键帧细化到阶段、时刻和短片段，经只读工具暴露。

方法与数据

规划前读取粗层次，执行时递归进入并仅加载当前子目标片段；每任务一个演示即可。
基于RPent，LIBERO-PRO成功率从92.6%升至96.5%，LIBERO-Plus从86.7%升至95.8%。

价值判断

值得关注：无需训练、单演示即可显著提升成功率。
可复用点：层次化视频递归导航与按需加载机制。
局限/待核查：摘要未明确真实机器人或其他基准表现。

摘要

LLM agents that orchestrate frozen vision-language-action (VLA) policies improve across episodes through text memory, which records what the agent did but not how the task is done. A demonstration video shows it, but fits poorly into an agent's context. The full video slows every turn, fixed keyframes lose the contact detail that decides whether a grasp holds, and what the agent needs shifts from the task's structure while planning to the frames around each contact. We introduce Recursive Video In-Context Learning (RV-ICL), a training-free method that turns a demonstration into a hierarchy the agent navigates rather than a prompt it receives. The hierarchy is built from the sub-events of the demonstration, such as grasps and releases. Its levels grow finer, from keyframes of the whole task to phases, moments and short clips, and are exposed through read-only tools. The agent reads the coarse levels before planning. During execution it re-enters the hierarchy whenever a step needs more detail and loads only the clip of its current sub-goal. One demonstration per task is enough. Built on RPent, RV-ICL raises success from 92.6% to 96.5% on LIBERO-PRO and from 86.7% to 95.8% on LIBERO-Plus.

2. MemPilot: Orchestrating On-Demand Multimodal Memory Curation for LLM Agents

来源：arXiv
日期：2026-10-05
作者/机构：Haozhen Zhang, Haodong Yue, Quanyu Long, Jianzhu Bao, Qingyuan Liu, Tao Feng
主题标签：AI-Agent, arXiv
推荐等级：高
分类：cs.CL, cs.AI, cs.LG

一句话结论

MemPilot 提出了一种灵活的框架，通过强化学习优化多步 LLM 策略，在不同性能-成本-延迟偏好下，对接需内存 curation，实现 LLM 代理的 on-demand 多模态内存管理。该框架解决了现有系统在构建内存时不必要的预处理成本和可能丢失关键细节的问题，并提供了灵活的运行时适应控制。

核心内容

Memory has become integral to the LLM agent ecosystem, supporting information retention and reuse across interactions, but most existing systems construct memory in a query-agnostic manner, incurring unnecessary preprocessing cost and discarding details that later prove essential.
Recent studies have begun shifting memory processing toward runtime adaptation, but typically specialize in particular operations or fixed processing schemes, leaving flexible control over performance, cost, and latency largely underexplored.
MemPilot orchestrates on-demand memory curation under different performance-cost-latency preferences by optimizing a multi-step LLM policy via reinforcement learning to iteratively choose between retrieving from query-agnostic memory and delegating query-specific curation of raw multimodal history to heterogeneous LLMs and VLMs.

方法与数据

优化 multi-step LLM policy via reinforcement learning，jointly controls evidence amount, curation instructions, model selection, and visual access。
采用 objective-wise advantage decoupling by separately estimating each objective’s advantage before aggregation，以及 prefix-based marginal utility estimation for fine-grained credit assignment across multi-step rollouts。
Experiments on five multimodal agent-memory benchmarks demonstrate favorable performance—cost—latency trade-offs。

价值判断

值得关注：MemPilot 提供了一种灵活的框架，在 LLM 代理中 orchestrates on-demand multimodal memory curation，实现性能-成本-延迟的精细权衡。
可复用点：通过 RL 优化多步策略，并使用 advantage decoupling 和 marginal utility estimation 来处理多目标优化。
局限/待核查：具体实验数据集和基准未明确，实验结果的具体数值和细节未提供。

摘要

Memory has become integral to the LLM agent ecosystem, supporting information retention and reuse across interactions. However, most existing agent memory systems construct memory in a query-agnostic manner, which can incur unnecessary preprocessing cost and discard details that later prove essential. Recent studies have begun shifting memory processing toward runtime adaptation, but typically specialize in particular operations or fixed processing schemes, leaving flexible control over performance, cost, and latency largely underexplored. To address this challenge, we present \textbf{MemPilot}, a flexible framework that orchestrates on-demand memory curation under different performance--cost--latency preferences. Specifically, we optimize a multi-step LLM policy via reinforcement learning to iteratively choose between retrieving from query-agnostic memory and delegating query-specific curation of raw multimodal history to heterogeneous LLMs and VLMs. The policy jointly controls evidence amount, curation instructions, model selection, and visual access, enabling fine-grained allocation of runtime computation. To optimize this policy under competing objectives, we adapt objective-wise advantage decoupling by separately estimating each objective's advantage before aggregation. Moreover, we introduce prefix-based marginal utility estimation for fine-grained credit assignment across multi-step rollouts. Experiments on five multimodal agent-memory benchmarks demonstrate favorable performance--cost--latency trade-offs across optimization preferences, with preference sweeps yielding broader frontiers than existing trade-off-aware baselines.