Hybrid position/force control is an effective tool for carrying out a task whose geometry constrains the position of a manipulator. This paper presents a task-oriented architecture for hybrid control, taking into consideration the complexity of the manipulator dynamics and uncertainties in the external constraints. For this purpose an adaptive pole assignment self-tuning algorithm was adopted based upon six independent decoupled ARMA models which represent position and force dynamics of manipulators in the task-oriented coordinate frame. To complete the control architecture, a control input transformation algorithm and an output synthesizing algorithm were developed in the task-oriented frame. These algorithms were designed to be easily transformable between coordinate frames in order to be applicable to a variety of tasks. To demonstrate the validity of the architecture, three example tasks were simulated using a manipulator whose kinematic and dynamic characteristics are analogous to those of a Puma 560 industrial robot.