AI-powered robot swarm shows 99.67% success in firefighting trial

A team led by Cyborg Dynamics Engineering and Griffith University in Australia successfully demonstrated AI-powered firefighting robots capable of navigating obstacles and collaboratively extinguishing fires without endangering human crews. Using an unmanned ground vehicle (UGV) paired with up to four virtual robot teammates, the system achieved a 99.67% success rate in simulated trials involving multiple fires. The robots were trained through a structured three-stage multi-agent reinforcement learning (MARL) curriculum, progressing from basic navigation to coordinated firefighting in complex environments. This approach enabled the robots to self-organize, divide tasks, and operate autonomously, reducing the cognitive burden on human operators and enhancing safety. The project builds on existing remotely controlled firefighting UGVs already deployed in Australian mine sites, with the next goal being full automation of low-level controls and swarm coordination. Autonomous swarms can respond more rapidly and effectively than human operators by leveraging sensor-driven decision-making and real-time data processing. The researchers plan to refine neural networks and improve