A Study on Adaptation of Gait Trajectory to Exoskeleton Robot for Paraplegic Patients Using the Learning Algorithm

Abstract

In order to let paraplegic patients walk with an exoskeleton robot, it is necessary to generate a walking pattern of the exoskeleton robot. The walking pattern of the exoskeleton robot was implemented by a single walking trajectory. In this way, the stride length can not be modified as much as you like. Therefore, there is a disadvantage in that it could be walked only in a certain pattern regardless of the change of the walking environment. In actual walking environment, there are many obstacles such as portholes and sidewalk curb, and in such a situation, it is necessary to walk with various strides in order to walk naturally. Also, when the user is in a state of falling forward, it is possible to prevent the user from falling down by increasing the stride length. Therefore, it is important to change the stride length in accordance with the intention or walking environment in activities of daily living (ADL). The purpose of this paper is to learn the walking pattern to the exoskeleton robot and to generate the walking patterns by the various stride length. First, derive the human's walking pattern using the vision sensor (Kinect Sensor). Obtained the walking trajectory is used as data to be learned to the robot. Second, the learning method adopts the Dynamic Movement Primitive (DMP) which is one of the trajectory control/planning methods. Through the DMP method, the exoskeleton robot learns the walking pattern of the human’s walking pattern and walks to the reconstructed patterns. The stride length in the walking pattern is modified according to the gain value. Finally, derive the proportional relation between the stride length output from the actual robot according to the gain value through experiment. By deriving the exact gain value of the required stride length, a gait pattern can be generated with a stride length that is optimized for the user's intention or gait environment.