In a distributed robotic system, both human-controlled (semi-autonomous) and computer-controlled (fully autonomous) robots may simultaneously exist. From the global system's point of view, supervisory control for the interactions between the human and computer are important and necessary. For such human-computer interactive systems, this paper proposes a supervisory framework to guarantee that both human and computer commands meet collision-free and deadlock-free requirements. In the presented approach, Petri nets are applied to construct a system model and synthesize a desired supervisor. An application to a two-robot remote surveillance system is provided to demonstrate the practicability of the developed supervisory control approach. It is believed that the technique developed in this paper is significant in the industrial practice.