Automatic Adaptation Of A Natural Language Interface To A Robotic System

This paper shows an application of four neural networks architectures for the automatic adaptation of the voice interface to a robotic system. These architectures are flexible enough to allow a non-specialist user to train the interface to recognize the syntax of new commands to the teleoperated environment. The system has been tested in a real experimental robotic system applied to perform simple assembly tasks, and the experiments have shown that the networks are robust and efficient for the trained tasks.