Quantum computing and neuromorphic computing for safe, reliable, and explainable multi-agent reinforcement learning: optimal control in autonomous robotics

Quantum computing and neuromorphic computing for safe, reliable, and explainable multi-agent reinforcement learning: optimal control in autonomous robotics

연구 분야: Artificial Intelligence

학회: Iran Journal of Computer Science

This paper introduces a novel hybrid framework that integrates quantum computing and neuromorphic computing to enhance the safety, reliability, and explainability of Multi-Agent Reinforcement Learning (MARL) in autonomous robotic systems. The proposed architecture employs quantum variational circuits for high-level policy exploration and spiking neural networks for energy-efficient, low-latency motor control. Adopting a centralized training and decentralized execution paradigm, the framework enables agents to optimize joint policies that combine quantum planning with neuromorphic execution under partial observability and safety constraints. We evaluate the framework in a simulated environment featuring ten UAV agents navigating dynamic forest terrain with limited visibility and obstacle avoidance requirements. Empirical results demonstrate that the hybrid system significantly reduces safety violations while maintaining entropy-based exploration and interpretable spike-based decision traces. KL divergence metrics confirm the convergence of quantum policies toward safe priors, while spike entropy analysis reveals temporal diversity in control signals. The key contributions of this work include: (i) a modular quantum-neuromorphic MARL architecture, (ii) a hybrid training framework incorporating safety-aware coordination, and (iii) empirical validation through both visual diagnostics and formal metrics. This research establishes a foundation for next-generation embodied AI systems that unify the optimization capabilities of quantum computing with the biological plausibility of neuromorphic control.