The paper presents a control architecture for robotic-assisted surgery in the presence of physiological motions. Dynamic and kinematic models, operational space, computed torque, discrete state space and stochastic design are addressed in the control. Inner loops are based on position and velocity signals and outer loops have force measurements. Two active observers (AOBs) are introduced for force control and motion compensation. The first AOB is responsible for model-reference adaptive control to guarantee a desired closed loop dynamics for the force. The second AOB performs control actions to compensate physiological motions. Such motions are described by a second-order stochastic equation, without apriori knowledge of signal characteristics. Simulation results are presented for sinusoidal and non-sinusoidal motions, highlighting merits of the approach.