Motion generation with a recurrent neural network

Several techniques can be used to generate motion sequences in computer animation. This paper explores the use of a recurrent neural network as an alternative way to create these sequences. It is found that this architecture can easily learn to produce a variety of linear and circular motion sequences and can also combine several of these sequences to produce smooth movements. More general sequences can also be learned and reproduced, although with somewhat more difficulty. Motion generation can also be guided by a plan. The incorporation of a plan enables accurate production of composite motion sequences