Primitive action learning using fuzzy neural networks

The learning of primitive actions, or affordances as often called, has always been one of the top items in the research agenda of the robotics community. In this paper we propose fuzzy neural networks as a viable solution for their computational efficiency, their ability to approximate smooth non-linear functions and their transparency of the underlying mechanisms of the trained network. More specifically we benchmark the Takaki-Sugeno Fuzzy Neural Network (TSFNN) in an experimental scenario where the robot learns to control its arm velocity to push a rolling object in a requested position. The experimental scenario was kept simple and of linear nature in order to benchmark the TSFNN with a least squares linear model. The real time experiments using a PR2 robot have been conducted to verify the proposed method. The experimental results have shown that the TSFNN is able to reliably and robustly learn and demonstrate the pushing action.