RSMeM: Knowledge-Enhanced Memory Evolution for Remote Sensing Agents with Systematic Evaluation
Quick Answer
RSMeM introduces a knowledge-enhanced memory evolution mechanism for remote sensing agents, improving tool-use performance by 6% on DeepSeek-V3.2 with less than 1% additional experience tokens.
Quick Take
This system integrates hierarchical knowledge grounding and failure-aware experience refinement to enhance domain-specific workflows, addressing the limitations of general-purpose in geoscience applications.
Key Points
- RSMeM consists of hierarchical knowledge grounding and failure-aware experience refinement.
- It improves end-to-end answer accuracy across diverse LLM backbones.
- Extensive experiments on EarthBench validate RSMeM's effectiveness.
- The system distills failure-annotated traces into reusable constraints.
- Code for RSMeM is publicly available for further research.
DeepSignal Analysis
What happened
RSMeM is a new mechanism designed to enhance the performance of remote sensing agents by integrating domain-specific knowledge. It has shown a 6% improvement in tool-use performance on the DeepSeek-V3.2 benchmark while requiring less than 1% additional experience tokens. This approach addresses the limitations of general-purpose LLMs in geoscience applications.
Key evidence
- RSMeM improves tool-use performance by 6% on the DeepSeek-V3.2 benchmark with less than 1% additional experience tokens.
- The mechanism consists of two components: Hierarchical Knowledge Grounding and Failure-Aware Experience Refinement.
- Extensive experiments on EarthBench demonstrate consistent improvements in tool-use performance across various LLM backbones.
Why it matters
The introduction of RSMeM is significant as it addresses the shortcomings of existing remote sensing agents that rely on general-purpose LLMs, which often lead to unreliable workflows. By incorporating domain-specific knowledge and refining experiences based on failures, RSMeM aims to create more robust and effective tools for geoscience research, potentially leading to more accurate analyses and insights.
Paper Resources
📖 Reader Mode
~2 min readAuthors:Bingxian Wu, Yu Zhang, Zonghao Guo, Tang Liu, Chen Qian, Yuxiang Lu, Xingbo Du, Yanghao Li, Yidan Zhang, Chi Chen, Ling Yao, Maosong Sun
Abstract:Geoscience research requires complex analysis and domain expertise, with remote sensing (RS) observations as a key foundation. However, existing RS agents built on general-purpose LLMs remain largely domain-agnostic, resulting in brittle and error-prone workflows. Moreover, these failures are seldom consolidated into a reusable experience for subsequent analyses. To address this issue, we introduce RSMeM, a knowledge-enhanced memory evolution mechanism that bootstraps RS agents with pre-distilled domain knowledge and iteratively integrates online experience for robust multi-step tool execution. RSMeM is composed of two components: (i) Hierarchical Knowledge Grounding, which performs taxonomy-aware retrieval over a hierarchical domain corpus to guide planning and tool selection; and (ii) Failure-Aware Experience Refinement, which distills failure-annotated tool-use traces into reusable constraints for next-round tool execution. By iteratively employing these two processes, RS agents can evolve to absorb task-level domain knowledge and effectively translate it into instance-level execution experience. Extensive experiments on EarthBench demonstrate that RSMeM consistently improves tool-use performance and end-to-end answer across a diverse set of LLM backbones. Notably, RSMeM achieves a 6% accuracy improvement on DeepSeek-V3.2 with less than 1% additional experience tokens, demonstrating the strong knowledge density of our distilled experience. Our code is available at this https URL
| Comments: | Accepted to ACL 2026 Main. Code: this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| ACM classes: | I.2.7; I.2.11 |
| Cite as: | arXiv:2607.24772 [cs.AI] |
| (or arXiv:2607.24772v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.24772 arXiv-issued DOI via DataCite |
Submission history
From: Bingxian Wu [view email]
[v1]
Thu, 11 Jun 2026 04:23:18 UTC (1,473 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising
HOBA (Hierarchical On-policy Bidding Agents) is a novel hierarchical reinforcement learning framework that enhances online advertising bidding systems by improving adaptability and reducing hyperparameter tuning costs. It utilizes a for hyperparameter inference, a SARSA agent for expert model selection, and a dynamic expert pool for bid execution, achieving a +3.6% increase in target cost during large-scale deployment and outperforming state-of-the-art baselines on AuctionNet.