Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Agent Geospatial Copilots for Remote Sensing Workflows

Chaehong Lee, Varatheepan Paramanayakam, Andreas Karatzas, Yanan Jian, Michael Fore, Heming Liao, Fuxun Yu, Ruopu Li, Iraklis Anagnostopoulos, Dimitrios Stamoulis

TL;DR

This paper addresses scalability challenges in geospatial remote sensing copilots by introducing GeoLLM-Squad, a multi-agent framework that decouples orchestration from geospatial task solving. It leverages AutoGen and GeoLLM-Engine to orchestrate diverse geospatial agents across urban, forestry, climate, agriculture, and satellite-vision tasks, achieving up to 17% improvements in agentic correctness over state-of-the-art baselines. The approach combines composition-based reasoning with iterative ledger reassessment and employs a large set of APIs (521 functions) to support robust, scalable workflows. Results show GeoLLM-Squad outperforms single-agent and ledger-based multi-agent systems, while remaining scalable with open-source LLMs such as Qwen-2.5 and GPT-4o-mini. The work highlights the potential of multi-agent AI to advance RS workflows while offering practical avenues for cost-aware deployment.

Abstract

We present GeoLLM-Squad, a geospatial Copilot that introduces the novel multi-agent paradigm to remote sensing (RS) workflows. Unlike existing single-agent approaches that rely on monolithic large language models (LLM), GeoLLM-Squad separates agentic orchestration from geospatial task-solving, by delegating RS tasks to specialized sub-agents. Built on the open-source AutoGen and GeoLLM-Engine frameworks, our work enables the modular integration of diverse applications, spanning urban monitoring, forestry protection, climate analysis, and agriculture studies. Our results demonstrate that while single-agent systems struggle to scale with increasing RS task complexity, GeoLLM-Squad maintains robust performance, achieving a 17% improvement in agentic correctness over state-of-the-art baselines. Our findings highlight the potential of multi-agent AI in advancing RS workflows.

Multi-Agent Geospatial Copilots for Remote Sensing Workflows

TL;DR

This paper addresses scalability challenges in geospatial remote sensing copilots by introducing GeoLLM-Squad, a multi-agent framework that decouples orchestration from geospatial task solving. It leverages AutoGen and GeoLLM-Engine to orchestrate diverse geospatial agents across urban, forestry, climate, agriculture, and satellite-vision tasks, achieving up to 17% improvements in agentic correctness over state-of-the-art baselines. The approach combines composition-based reasoning with iterative ledger reassessment and employs a large set of APIs (521 functions) to support robust, scalable workflows. Results show GeoLLM-Squad outperforms single-agent and ledger-based multi-agent systems, while remaining scalable with open-source LLMs such as Qwen-2.5 and GPT-4o-mini. The work highlights the potential of multi-agent AI to advance RS workflows while offering practical avenues for cost-aware deployment.

Abstract

We present GeoLLM-Squad, a geospatial Copilot that introduces the novel multi-agent paradigm to remote sensing (RS) workflows. Unlike existing single-agent approaches that rely on monolithic large language models (LLM), GeoLLM-Squad separates agentic orchestration from geospatial task-solving, by delegating RS tasks to specialized sub-agents. Built on the open-source AutoGen and GeoLLM-Engine frameworks, our work enables the modular integration of diverse applications, spanning urban monitoring, forestry protection, climate analysis, and agriculture studies. Our results demonstrate that while single-agent systems struggle to scale with increasing RS task complexity, GeoLLM-Squad maintains robust performance, achieving a 17% improvement in agentic correctness over state-of-the-art baselines. Our findings highlight the potential of multi-agent AI in advancing RS workflows.

Paper Structure

This paper contains 6 sections, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background and Novelty
  3. Methodology
  4. Results
  5. Future Work and Opportunities
  6. Conclusion

Figures (2)

  • Figure 1: GeoLLM-Squad is a multi-agent Copilot for RS workflows that builds on the open-source AutoGen wu2023autogen and GeoLLM-Engine Singh_2024_CVPR frameworks. Our orchestrator supports various geospatial agents and API tools to handle diverse workflows such as climate analysis or urban planning tasks.
  • Figure 2: Single- vs. multi-agent scalability ablations: we evaluate each method (with TS paramanayakam2024less and WM wang2024agent incorporated) separately for each task and for multi-task combinations. GeoLLM-Squad consistently provides better agentic performance across all tasks.