Associative motion generation for humanoid robots based on analogy with indication

Abstract

We describe a method of generating new motions associatively from unfamiliar indications. The associative motion generation system is composed of two neural networks: nonlinear principal component analysis (NLPCA) and Jordan recurrent neural network (JRNN). First, the system learns the correspondence relationship between an indication and a motion using training data. Second, associative values are extracted for associating a new motion from an unfamiliar indication using NLPCA. Last, the robot generates a new motion through calculation by JRNN using the associative values. Experimental results demonstrated that our method enabled a humanoid robot, KHR-2HV, to associatively generate some kinds of motion depending on given unfamiliar indications.