Stacked hidden Markov model for motion intention recognition

Motion intention recognition plays an important role in robot-assisted applications. A Stacked Hidden Markov Model (HMM) method is proposed to enable the robot to recognize the intention of a human user based on his/her motion trajectories. The Stacked HMM method is constructed based on the relationship of the observed objects. The motion intention recognition model contains multiple HMMs. Each HMM represents one motion intention in the corresponding level. Motion trajectories were collected from a Virtual Reality based surgical training platform. A two-Layered Stacked HMM intention recognition model has been built to recognize the motion intention in primitive level and subtask level. With the proposed intention recognition method, intention recognition rate for the primitive and subtask levels are 95.0±3.5% and 71.0±13.6% respectively. The proposed method is effective in the recognition of user's intention from different levels with motion trajectory.