We propose an approach for modeling visually guided behaviors of agents which explore and navigate in unknown and partially known environments. Behaviors are modeled as finite state machines (FSM), where the states of the model correspond to particular continuous control strategies and the transitions between them are caused by events representing qualitative or asynchronous changes in the behavior evolution. In order to prevent conflicts in parallel execution of multiple behaviors we adopt the supervisory control theory of discrete Event System (DES). Modeling the participating processes using the DES framework allows us to capture often complex interactions between components of the system and synthesize the resulting supervisor, guaranteeing the overall controllability of the system at the discrete event level. In the real world agents have multiple options/paths for carrying out their task. Hence there is a need for selecting different control strategies based on efficiency and safety criteria. We have included in our formalism a measure of efficiency as the nominal cost (in our case, the traversal time) and a measure of safety at the risk cost (in our case, the inverse of the distance between the agent and obstacles). Experiments have been carried out testing the described formalism with one agent carrying out the task of avoiding an obstacle in its path while tracking a target.