Motion reproduction by human demonstration based on discrete hidden Markov model for nursing-care assistant robot

Generating motions for nursing-care assistant robot in hospital environment is a challenge. The cost of teaching an unexperienced user to operate the robot skillfully is very high. In this research a Discrete HMM (DHMM) is trained on samples of skilled operator and then the robot motions is reproduced by the DHMM and related algorithms generating optimum state sequence, Key Motion Frames (represented by discrete output symbols) and interpolated robot joint trajectories. The proposed motion generation method has a simple framework which allows the engineer customize the trajectories by editing key motion frames according to different requirements of tasks. This endows the robot the ability that utilize the skills of experienced operator adaptively in similar situations.