<p>In this work, the intrawave frequency modulation is investigated for robotic milling under different cutting states. The dynamic response of robotic milling is modeled by multiple multi-sinusoidal chirplet modes without overlapping, while noise and harmonic interferences are considered. An improved chirplet transform method is developed to identify the fast-varying intrawave frequency modulation under strong interferences. Firstly, the proposed method uses time-dependent windows, so high-resolution time-frequency analysis can be realized through iterative updating of both windows and chirplet kernels. In the time-frequency domain, the noise energy can be greatly suppressed by applying the maximum window length for all time nodes of the signal. Secondly, the posteriori estimate of peak frequencies measured from the time-frequency distribution is calculated by the robust Kalman filter, and then approximated by a local fitting tool of the moving least square. Estimation errors caused by harmonic interferences and overfitting can be effectively reduced. Through numerical simulation, a high accuracy of the developed method is validated. In the cutting tests, it is found that the center frequencies of intrawave frequency modulations are located at the tooth passing frequency and its integer multiples, while the sine waves of intrawave frequency modulations show various patterns with respect to different cutting states. Based on the estimated intrawave frequency modulation, a new chatter indicator is extracted and can alert the chatter presence much earlier than chatter marks appear on the workpiece.</p>