This paper presents a multi-agent infrastructure that couples physical and digital robotic environments, enhanced by a knowledge graph. The infrastructure aims to simplify the design, modification, and evaluation of robotic systems. We propose to use a knowledge graph to represent environmental data, robot states, and interactions. This enables supervision of both physical and simulated robots on existing and heterogeneous platforms. Deployed in the Hypermedea multi-agent programming environment, agents use the knowledge graph to facilitate seamless interaction and coordination across different platforms. The feasibility of the infrastructure is demonstrated through a coordination scenario between a simulated robotic arm and a physical mobile robot in Industry 4.0.