Motion Generation of Humanoid Robot based on Polynomials Generated by Recurrent Neural Network

摘要

Humanoid robots are expected to have a variety of motions enabling them to interact with people on a sophisticated level. Making a program for generating several stable motions using standard programming languages such as C, is not only time consuming but is also difficult to understand and modify. For this, a suitable recurrent neural network language (RNN) inspired by neurobiology has been developed. In this paper, a simple method of motion generation based on polynomials generated by RNN is presented. All motions are generated using a basic RNN circuit of a first order polynomial. This method easily enables a humanoid robot to execute complex motions. Furthermore, feedback controllers can be easily inserted in the RNN circuit of a motion with any desired timing. Both rhythmic and non-rhythmic motion can be generated based on the same strategy. The effectiveness of the proposed method is verified by experimental results.