GeoDisaster: Benchmarking Orchestrated Agents for Operational Disaster Geo-Intelligence
Researchers have introduced GeoDisaster, a new benchmark designed to evaluate and improve the capabilities of orchestrated agents in operational disaster geo-intelligence. This benchmark includes 2,921 instances across five task families, integrating diverse Earth observation and GIS data for tasks like hazard detection and damage assessment. The accompanying multi-agent framework utilizes a novel alignment technique called Role-Contract Expectation Alignment (RCEA) to enhance tool use and decision-making in disaster response scenarios. AI
IMPACT This benchmark could drive advancements in AI agent capabilities for real-world applications like disaster response and geo-intelligence.