We present a framework for imitation of percussion performances with parameter-based learning for accurate reproduction. We constructed a robotic setup involving pull-solenoids attached to drum sticks which communicate with a computer through an Arduino microcontroller. The imitation framework allows for parameter adaptation to different mechanical constructions by learning the capabilities of the overall system being used. For the rhythmic vocabulary, we have considered regular stroke, flam and drag styles. A learning and calibration system was developed to efficiently perform grace notes for the drag rudiment as well as the single stroke and the flam rudiment. A second pre-performance process is introduced to minimize the latency difference between individual drum sticks in our mechanical setup. We also developed an off-line onset detection method to reliably recognize onsets from the microphone input. Once these pre-performance steps are taken, our setup will then listen to a human drummer’s performance pattern, analyze for onsets, loudness, and rudiment pattern, and then play back using the learned parameters for the particular system. We conducted three different evaluations of our constructed system.