Abstract The rapid evolution of smart manufacturing systems has fundamentally transformed the operational paradigm of industrial robotics. Unlike traditional automation frameworks characterized by rigid, centralized programming, modern production environments require adaptive, cooperative robotic agents capable of decentralized decision-making and dynamic task redistribution. As production lines become increasingly modular and demand variability intensifies, robotic systems must exhibit autonomy not only at the motion control level but also at the coordination and strategic planning levels. This paper proposes a Multi-Agent Cooperative Control Architecture (MACCA) designed to enable distributed intelligence among heterogeneous industrial robots operating within Industry 4.0 environments. The proposed framework integrates graph-theoretic consensus control, deep reinforcement learning, distributed trajectory planning, and edge-based computational autonomy. Each robotic unit functions as an intelligent agent equipped with local sensing, onboard processing capabilities, and peer-to-peer communication modules. The architecture introduces a hierarchical yet decentralized control mechanism in which low-level motor control remains embedded within robotic platforms, mid-level coordination emerges through distributed negotiation protocols, and high-level optimization is performed through supervisory analytics without imposing centralized control authority. The system is evaluated through a simulated smart factory environment incorporating autonomous mobile robots, articulated manipulators, and collaborative robotic arms. Experimental results demonstrate significant improvements in operational efficiency, collision mitigation, and energy optimization under dynamic production loads. The proposed architecture contributes to applied robotics engineering by offering a scalable and fault-tolerant cooperative framework capable of sustaining industrial-grade real-time performance while reducing dependence on centralized infrastructure.
Kwon et al. (Tue,) studied this question.