Estimation of stochastic representation of via-points in human motion control by reinforcement learning

Humans can generate a complex trajectory by imitating the movements of others. A method for learning complex sequential movements that utilizes via-point representation is proposed. However, the proposed algorithm for estimating a set of via-points from complex movement does not involve such a learning process as trial and error. Instead, it finds the minimum number of via-points and then specifies a unique set of them without a trial and error process. In this paper, we report a learning algorithm for stochastic viapoint representation through trial and error in a human-like manner. Based on reinforcement learning, the proposed viapoint algorithm locates a set of via points that mimics reference trajectory by iterative learning and uses the evaluation values of a generated movement pattern.