EarthVision Agents - Earth Vision AI Alliance

Overview

EarthVision Agents represents a paradigm shift in how we approach complex Earth Observation (EO) tasks. Rather than building single-purpose AI models, we develop autonomous agent frameworks that can reason, plan, and execute multi-step workflows to solve sophisticated remote sensing challenges.

These agents combine vision-language understanding, tool use capabilities, and structured reasoning to tackle real-world EO problems that require breaking down complex queries into manageable steps, utilizing specialized tools, and synthesizing information across multiple data sources. Our agent frameworks enable more flexible, interpretable, and powerful Earth AI systems that can adapt to diverse and evolving analysis requirements.

Key Capabilities

🧠

Structured Reasoning

Agents break down complex EO tasks into logical steps, planning and executing multi-stage analysis workflows with clear reasoning chains.

🔧

Tool Augmentation

Integration with specialized remote sensing tools and APIs, enabling agents to leverage domain-specific capabilities for advanced analysis.

🎯

Task Decomposition

Automatic decomposition of high-level user queries into executable sub-tasks, managing dependencies and orchestrating complex workflows.

🔄

Adaptive Problem Solving

Dynamic adaptation to task requirements, selecting appropriate strategies and tools based on the specific EO analysis context.

Research Projects

ThinkGeo

Project Page arXiv

ThinkGeo is a specialized benchmark designed to evaluate how language model agents handle complex remote sensing tasks through structured tool use and step-by-step reasoning. The benchmark assesses agent capabilities across diverse EO scenarios requiring planning, tool selection, and multi-step problem decomposition. ThinkGeo provides comprehensive evaluation metrics for measuring agent performance in real-world Earth observation applications, helping advance the development of more capable and reliable autonomous EO systems.

[arXiv 2025] Shabbir, Akashah, et al. "ThinkGeo: Evaluating Tool-Augmented Agents for Remote Sensing Tasks." arXiv preprint arXiv:2505.23752 (2025).