A severity-aware multi-agent soft actor-critic framework for medical resource coordination in plateau disaster rescue

Area of Science:

• Artificial Intelligence
• Robotics
• Emergency Medicine
• Environmental Physiology

Background:

• High-altitude autonomous medical rescue is challenged by extreme environments (hypoxia, cold, terrain) and limited infrastructure.
• Existing multi-agent reinforcement learning (MARL) struggles with integrating discrete maneuvers and continuous medical resource allocation.
• The critical
• platinum 10 min
• window necessitates rapid, precise decision-making under severe constraints.

Purpose of the Study:

• To develop an advanced AI framework for autonomous medical rescue in high-altitude environments.
• To overcome limitations of current MARL approaches in balancing tactical movement and precise medical resource management.
• To enhance survival rates and logistical efficiency in extreme, resource-scarce rescue scenarios.

Main Methods:

• Proposed ISMRG² mSAC (injury severity-medical resource collaborative guided multi-agent soft actor-critic), a novel MARL framework.
• Integrated a hybrid discrete-continuous action space and a severity-aware attention mechanism.
• Utilized a maximum entropy objective for enhanced exploration and a "micro-dosing" strategy for adaptive drug administration.

Main Results:

• Achieved an 84.2% survival rate and a 0.83 drug efficiency ratio, significantly outperforming baseline methods.
• Demonstrated effective balancing of operational agility and safety in rescue missions.
• Showcased superior mission completion times and logistical sustainability compared to existing approaches.

Conclusions:

• The ISMRG² mSAC framework successfully integrates discrete maneuver control with continuous medical resource allocation for high-altitude rescue.
• The severity-aware attention mechanism and hybrid action space enable adaptive, fine-grained treatment decisions.
• This approach yields substantial improvements in casualty survival and drug utilization efficiency.