T. Mukai, S. Kuriyama, and T. Kaneko (Japan)
Human animation, Hierarchical learning, Motion data
This paper proposes a method for creating human move ments by imposing positional constraints of end-effectors at multiple key-frames. We use hierarchical reinforcement learning for efficiently extracting features of motion cap ture data in order to search postures at each key-frame among the huge number of possible candidates. The me chanical structures of virtual humans are also hierarchi cally decomposed so as to suit the learning mechanism for narrowing down the searching space, and reward func tions are also designed so as to reflect such a hierarchy. Our method automatically generates complicated sequen tial movements of the whole body from multiple con straints on end-effectors so that the resulting motions in herit the naturalness involved in motion capture data.
Important Links:
Go Back
