A recent development for programming industrial robotic manipulators is the interactive teaching of task motions where a human operator physically guides the end‐effector of a robot. Admittance control schemes are used to this end. Since the operator is in direct contact with the robot, it must be ensured that the velocities of the robot are within a safe range at any time. Thus, in order to guarantee safe operation, position and velocity limits must be respected within the admittance control. Existing admittance control schemes can either take task space limits or joint space limits into account. In this paper an extended admittance control scheme is presented which is capable of simultaneously taking task space limits as well as joint space limits into account. This is achieved by formulating the admittance control law in terms of a path parameter to account for task space limits and a subsequent determination of the path parameter dynamics with respect to the joint space limits. Admittance control requires an estimation of the interaction wrench to be available. Three different methods for obtaining such an estimation are presented: 1) A six‐axis force/torque sensor mounted on the robot end‐effector, 2) The generalized momentum, 3) A decoupled mechanical model of the motors. The performance of the algorithms is demonstrated by means of experimental results.