Haptic devices are now being deployed in several applications. One of the most interesting is the bilateral telerobotics, as such devices can provide a vivid tactile and/or proprioceptive rendering of the remote environment. In order to make this type of devices more affordable, low cost technology is going to be used, with a possible reduction in performance, due to the loss of backdriveability, especially when sensorless force control is implemented. The use of Reaction Force Observers (RFOB) can ameliorate the overall performance, but such solution achieves a good performance only when an accurate model of the electromechenical system is available. This paper proposes the use of an Acceleration-aided Kalman Filter (aaKF) to implement an accurate identification of the mechanical parameters of the haptic device. Experimental results show how the aaKF-based RFOB, with a properly designed identification procedure of the non-linear friction parameters, can match in performance a force control with an actual force sensor.