A Study of Gait Terrain Distinction AlgorithmUsing Rule-based Fuzzy Inference Systemfor Smarter Wearable Robot

This Paper proposed a gait terrain distinction algorithm for the walking of exoskeleton robot. To operate in the real environment wearing exoskeleton robot, user can walk not only on the level ground, but also ascent or descent the stair and ramp environment. Because gait phase and dynamic model of robot should be changed continuously while walking of the different terrain, terrains like stair and ramp need to be distinguished from level ground. Based on distinguished terrains, robot can be controlled to achieve stability while walking. To recognize current gait terrain, Inertial Measurement Unit(IMU) sensors and insole sensors are used. IMU sensors are attached to the lateral side of thigh and ankle to get angles of each joint at the sagittal plane. Insole sensors are used to recognize ground contact and identify gait phase. Fuzzy inference system is implied to identify 5 terrain - level ground, stair ascent/descent, ramp ascent/descent -. Based on IMU sensor data and gait phase, 3 linguistic variables - terrain gradient, the highest swing thigh angle, thigh angle difference - are obtained and used to identify the degree of each membership variable. Trapezoidal function and left/right shoulder function are used as membership function to convert crisp input into linguistic variable. By using If-Then statement rules in inference process, linguistic variable is converted into output membership function. Triangle function is used in defuzzification step as a output membership function. In the defuzzification step, mamdani���s min-max method is used. To verify the performance of algorithm, experiments are proceeded for 5 healthy men in 20s. Subjects asked to walk on level ground, stair, and ramp as their preferred speed. Evaluation are proceed using comparison accuracy of terrain identification result on different terrain. Confusion matrix and average identification accuracy are used to verify performance of algorithm.Maste