Fast and sensorless peg-in-hole insertion is a challenging task for a robotic manipulator. In order to deal with the peg-in-hole insertion problem without any need of an external force/torque sensor, this paper proposes to actively accomplish compliance in the insertion task relying on an admittance based control. This is combined with a real-time trajectory generator, by means of constraint based optimization, where a model-based sensorless observer of interaction forces is exploited. Experiments have been performed on an ABB dual-arm 7-DOF lightweight prototype robot to validate the proposed approach, with an insertion speed comparable to human manual execution and in presence of geometric uncertainty.