Impact sound occurs when a collision between objects occurs. Using this impact sound, we propose a robotic control method to reproduce the desired impact sound, namely, the audio-feedback control for impact sound. It is rare to control the manipulator using sound information, but is very useful in our life. Therefore, we hope it has the possibility to control the manipulator in wider tasks. Impact sound occurs in very fast phenomena, so it is effective to use a learning control. In this paper, we propose a learning control method to optimize the least-squares criterion of learning error, and carry out experiments on audio-feedback control for the impact sound, showing the effectiveness of the proposed method. This method is applied to obtain a unique control gain based on the system's impulse response model. Therefore, if the impulse response model becomes available, the learning control system can be constructed without knowing the system's parameters.
