Balancing exploration and exploitation is a central challenge in multi-agent
reinforcement learning (MARL), particularly in dynamic environments where
uncertainty and partial observability complicate coordination among agents. This
paper introduces Q-CMAPO, a quantum-classical framework designed to address
this challenge by integrating variational quantum optimization with classical
policy optimization. The framework leverages quantum-inspired combinatorial
search to guide exploration in high-dimensional action spaces, while classical
control-theoretic principles ensure stability and convergence of decentralized policies.
Within this architecture, centralized training with decentralized execution
enables scalable coordination across heterogeneous agent teams, while graphtheoretic
representations support connectivity and information sharing under
partial observability.
The proposed methodology provides a principled mechanism for adaptively balancing
exploration and exploitation through quantum-classical synergy. Experimental
evaluations in UAV coordination tasks demonstrate improvements in
convergence rate, reward accumulation, and policy diversity under uncertain
dynamics. The design emphasizes both mathematical rigor and computational
tractability, ensuring that the framework can be extended beyond UAV scenarios
to more general MARL settings. To support transparency and reproducibility, the
full implementation of Q-CMAPO is made available in an open-source repository.