In this work we present the smart knee prosthesis designed for in-vivo kinematics measurement and its validation in two knee simulators, i.e. a robotic knee simulator to provide realistic condition, and a manual simulator with more degrees of freedom. The sensor configuration including three magnetic sensors was designed, and the machine learning techniques were used to translate the magnetic measurements to knee rotations. First the concurrent flexion-extension and internal-external rotations were estimated via linear and nonlinear estimators, and technically validated in a manual knee simulator against motion capture system. Then the flexion-extension estimation was validated in a robotic knee simulator providing the realistic sagittal kinematics of treadmill and over-ground walking. The obtained results showed the high accuracy and precision of the estimates.