Comparison of sensor fusion methods for an SMA-based hexapod biomimetic robot

This paper compares the performances of different sensor fusion algorithms in a shape memory alloy (SMA)-based hexapod biomimetic robot, SMABOT IV. The algorithms considered include a Kalman filter that minimizes the estimation error variance, an H∞H∞ filter that minimizes the worst-case estimation error, and a robust mixed Kalman/H∞H∞ filter that allows for uncertainties in both the system and measurement matrices. The sensors installed on the robot include an inertial measurement unit and an electric compass sensor for inertial guidance. In addition, a stride-length-estimation algorithm for an SMA-based legged robot was proposed to establish the legged odometry of the robot. Allan variance analysis is employed to identify the noise sources of inertial sensors, and the calculated variance values are used to design the parameters of the Kalman filter algorithm. Finally, experimental results of two-dimensional navigation are presented, and the performances of three sensor fusion algorithms are compared. The results indicate that after identifying the noise characteristics of inertial sensors, the Kalman filter provides the best performance.