InfernoBots

Our project is an AI-powered autonomous emergency response system built entirely in simulation, demonstrating how intelligent multi-robot teams can detect, assess, and respond to wildfire and disaster scenarios with least human intervention. The system integrates an aerial drone (DJI Mavic 2 Pro) and ground robots operating inside a realistic Webots simulation environment. The drone performs real-time visual surveillance using a Vision-Language Model (VLM) to detect fire, smoke, damaged vegetation, and injured humans. Instead of following pre-scripted paths, the drone uses a orchestrated agentic flow to reason about the scene, decide where to move next, prioritize threats, and select appropriate actions such as dropping water bombs to suppress flames or deploying first-aid kits and breathing masks for victims or navigating to next target around the scene of criticality. The ground robot supports logistics and follow-up operations, navigating autonomously to transport essential equipment based on the aerial assessment. A centralized supervisory AI coordinates task allocation, ensuring efficient parallel execution while avoiding collisions and redundant coverage. The system emphasizes adaptive decision-making, dynamic mission planning, and perception-driven autonomy. All behaviors are designed to be transferable to real robotic platforms, showcasing how AI reasoning combined with robotic control can enhance disaster response operations in real-world environments.