AI Extends Firefighter Breathing Time by a Third

Firefighters face a critical limitation in their gear: traditional breathing apparatuses waste oxygen and typically last only about 30 minutes under heavy exertion, forcing dangerous retreats from burning structures. A new system developed by Galactic Bioware aims to change that by integrating artificial intelligence directly into life support, creating a semi-closed-circuit breathing apparatus that recycles exhaled air. This approach could significantly extend mission times in environments where every minute counts, from structural fires to industrial accidents, by making every breath more efficient and adapting to the firefighter's needs in real time.

The researchers found that their AI control system improves operating endurance by 18% to 34% in simulations compared to conventional fixed-setpoint controllers. In one scenario with intermittent bursts of heavy exertion, the AI extended oxygen depletion time from 98 minutes to 131 minutes, a 33.7% increase. Crucially, it achieved this while maintaining tighter safety margins, such as keeping peak carbon dioxide levels below 0.49% compared to 0.72% with traditional control, and managing core temperature more effectively. The system also adheres to strict fire-safety limits, preventing oxygen enrichment above 23.5% in the suit atmosphere to reduce combustion risks in fiery environments.

Ology combines a rigorous physicochemical model with an AI architecture that fuses data from three sensor tiers. The physical system includes a soda lime scrubber to remove carbon dioxide, a silica gel dehumidifier to control humidity, and pure oxygen replenishment from a finite 3 kg tank, all within a positive-pressure suit that vents outward through one-way valves. The AI controller uses an 18-state nonlinear model, incorporating inputs from external environmental sensors (like radiant heat flux and toxic gas detectors), internal suit sensors (including triple-redundant oxygen cells with median voting), and firefighter biometrics (such as heart rate and activity from inertial measurement units). It employs a model-predictive control framework with a dynamic resource scarcity multiplier that increases penalty on waste as consumables deplete, a reinforcement learning policy advisor for warm-starting optimizations, and a final control-barrier-function safety filter to enforce hard constraints before any command reaches the hardware.

Simulation across three scenarios demonstrate the system's effectiveness. In a steady moderate exertion scenario, the AI controller extended oxygen depletion time from 142 minutes to 177 minutes, a 24.6% improvement, while keeping peak carbon dioxide at 0.43% versus 0.48% with baseline control. For an escalating thermal threat scenario with ambient temperature rising from 60°C to 300°C over 90 minutes, endurance increased by 18.7%, from 107 to 127 minutes, with better thermal management reducing peak core temperature. The AI's ability to anticipate and adapt to changing conditions, such as increasing fan speed for cooling or adjusting oxygen injection based on metabolic estimates, underpins these gains, showing how predictive optimization can outpace reactive control in dynamic, high-stakes environments.

Of this research extend beyond firefighting to any field where extended life support in hazardous conditions is needed, such as military operations, space exploration, or deep-sea diving. By making oxygen use more efficient, the system could reduce the weight firefighters carry, potentially improving mobility and reducing fatigue. The AI's integration of situational awareness—like thermal threat assessment from external sensors—allows for proactive resource management, helping firefighters stay safer longer in unpredictable scenarios. However, the study notes that current simulations assume a single continuous deployment; real-world applications will require addressing multi-sortie operations, scrubber cartridge replacement, and battery power constraints for field viability.

Limitations of the system include its reliance on simulations rather than physical prototypes, with assumptions that may not fully capture real-world complexities like sensor noise or extreme environmental variability. The paper highlights that the single-atmosphere architecture, where the entire suit volume is oxygen-enriched, poses a fire-safety due to rapid oxygen enrichment from venting; a separated breathing-loop design is suggested for future iterations to mitigate this. Additionally, the AI controller's computational feasibility on embedded hardware requires further validation, and cybersecurity risks in contested environments could introduce vulnerabilities that need hardening. The researchers also note that long-duration operations will require solutions for consumable replacement and power management, indicating that while the AI framework is promising, practical deployment hinges on addressing these engineering gaps.