Robot trajectory generation using modified hidden Markov model and Lloyd's algorithm in joint space

Human guide robots usually generate desired trajectories from human demonstrations. The training process can be in task space or joint space. The task space method needs the inverse kinematics; the joint space method uses dynamic time warping. Both of them destroy the accuracy of the generated trajectory.In this paper, we first use Lloyd's algorithm to encode the input signals such that the observations are time-independent. The desired trajectory is generated in joint space without dynamic time warping. Then we modify the hidden Markov model (HMM) such that it can work in joint space. Since the desired trajectories are the joint angles, they can be applied directly to robot control without calculating the inverse kinematics. Simulation and experimental results show that the modified HMM with Lloyd's algorithm work well in joint space.