Nonlinearity and uncertainty are the two most common features of real mechatronics systems. Therefore, the performance of extended Kalman filter(EKF), unscented-KF(UKF), and adaptive unscented-KF(AUKF) are compared by the extensive experiments on a multi-flying-robot test-bed, and with the same scenario of moving target tracking. The experimental results are analyzed with quantitation to show the effectiveness of the three KFs while facing nonlinearity and uncertainties.