Self-reflective segmentation of human bodily motions using associative neural networks towards human-machine shared autonomy

For realizing a naturalistic collaboration between humans and robots, we have to establish intention-sharing from the series of motion data that are observed and exchanged between the human and the machine. This is a problem of detecting meanings in the digitized data stream. We propose an approach based on semiosis, and present a number of ways for implementing the ideas using associative neural networks; one is a recurrent neural Elman network and the other one is Grossberg's adaptive resonance theory model. Experimental results are shown and their contributions to the design of a human-machine shared autonomy system are discussed.