Abstract: As millimeter wave radars (MMR) have been widely used in advanced driver assistance systems and autonomous driving vehicle to improve traffic safety, the radars may operate in a small road and in the same frequency band. Therefore, multi-path interference between different automotive radars arises, resulting in ghost targets. To deal with the emerging challenge of the multi-path interference, this paper presents an adaptive approach of MMR waveform optimization based on learning of multi-path interference. Firstly, the paper assumes that the multi-path interference follows a time-varying stochastic process; Thanks to an autoregressive model, the coefficients of the multi-path interference channels are dynamically modeled as a kind of finite impulse response; And then the parameters of the model of the neighboring multi-path interference are learnt by using a Kalman filter, based on sequential Bayesian inference. Secondly, the models of the desired radar signal and the multi-path interference are built. And the multi-path interference signals are whitening filtered. Finally, based on knowledge of the neighboring multi-path interference and a criteria of maximum signal to interference ratio (SIR), the anti-interference radar signal optimization model is generated, and therefore the MMR waveform is further optimized. The experimental results demonstrate that the proposed approach can significantly reduce neighboring multi-path interference, showing a promise for potential applications in automotive radar.