Tendon Driven System is widely used in various applications which requires compact and flexible moving parts. However, during the force transmission, TDS suffers unwanted disturbance like unknown friction, that causes inaccurate control performance. Therefore, there are several approaches to compensate the friction during the control, but it is hard due to the complex model of the internal friction. There are some known friction models that can represents the internal friction of the TDS, but depends on the system, it varies a lot. Therefore in this study, we estimated the internal friction model in real time by using the model-free machine learning algorithm. The main control was done by the PID controller but there is a friction compensation term based on the model-free machine learning algorithm. The algorithm keep updating the internal friction model that can reduce the estimation error, during the control process. In addition, the batch algorithm is used to reflect the overall motion trends and reduce the computation time. For the testbed of the controller, the simulation that represents the inverted pendulum with TDS was conducted. Since the algorithm is model-free, it could be applied to various TDS to increase the control performance.