MemCompiler: Compile, Don't Inject -- State-Conditioned Memory for Embodied Agents

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ding, Xin, Wang, Xinrui, Yang, Yifan, Wu, Hao, Jiang, Shiqi, Zhang, Qianxi, Mi, Liang, Zhu, Hanxin, Li, Kun, Liu, Yunxin, Chen, Zhibo, Cao, Ting
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909040921018368
author Ding, Xin
Wang, Xinrui
Yang, Yifan
Wu, Hao
Jiang, Shiqi
Zhang, Qianxi
Mi, Liang
Zhu, Hanxin
Li, Kun
Liu, Yunxin
Chen, Zhibo
Cao, Ting
author_facet Ding, Xin
Wang, Xinrui
Yang, Yifan
Wu, Hao
Jiang, Shiqi
Zhang, Qianxi
Mi, Liang
Zhu, Hanxin
Li, Kun
Liu, Yunxin
Chen, Zhibo
Cao, Ting
contents Existing memory systems for embodied agents typically inject retrieved memory as static context at episode start, a paradigm we term Ahead-of-time Monolithic Memory Injection (AMMI). However, this static design quickly becomes misaligned with the agent's evolving state and may degrade lightweight executors below the no-memory baseline. To address this, we propose MemCompiler, which reframes memory utilization as State-Conditioned Memory Compilation. A learned Memory Compiler reads a structured Brief State capturing the agent's current execution state and dynamically selects and compiles only relevant memory into executable guidance. This guidance is delivered through a text channel and a latent Soft-Mem channel that preserves perceptual information not expressible in text. Across Alf World, EmbodiedBench, and ScienceWorld, MemCompiler consistently improves over no-memory across open-source backbones (up to +129%), matches or approaches frontier closed-source systems, and reduces per-step latency by 60%, demonstrating that state-aware memory compilation improves both effectiveness and efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2605_07594
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MemCompiler: Compile, Don't Inject -- State-Conditioned Memory for Embodied Agents
Ding, Xin
Wang, Xinrui
Yang, Yifan
Wu, Hao
Jiang, Shiqi
Zhang, Qianxi
Mi, Liang
Zhu, Hanxin
Li, Kun
Liu, Yunxin
Chen, Zhibo
Cao, Ting
Robotics
Existing memory systems for embodied agents typically inject retrieved memory as static context at episode start, a paradigm we term Ahead-of-time Monolithic Memory Injection (AMMI). However, this static design quickly becomes misaligned with the agent's evolving state and may degrade lightweight executors below the no-memory baseline. To address this, we propose MemCompiler, which reframes memory utilization as State-Conditioned Memory Compilation. A learned Memory Compiler reads a structured Brief State capturing the agent's current execution state and dynamically selects and compiles only relevant memory into executable guidance. This guidance is delivered through a text channel and a latent Soft-Mem channel that preserves perceptual information not expressible in text. Across Alf World, EmbodiedBench, and ScienceWorld, MemCompiler consistently improves over no-memory across open-source backbones (up to +129%), matches or approaches frontier closed-source systems, and reduces per-step latency by 60%, demonstrating that state-aware memory compilation improves both effectiveness and efficiency.
title MemCompiler: Compile, Don't Inject -- State-Conditioned Memory for Embodied Agents
topic Robotics
url https://arxiv.org/abs/2605.07594