This paper presents an automatic parking for a passenger vehicle, with highlights on arobust real-time planning approach and on experimental results. We propose a frameworkthat leverages the strength of learning-based approaches for robustness to environmentsnoise and capability of dealing with challenging tasks, and rule-based approaches for itsversatility of handling normal tasks, by integrating simple rules with RL under a multi6 stage architecture, which is inspired by typical auto-parking template. By taking temporalinformation into consideration with using Long Short Term Memory (LSTM) network, ourapproach could facilitate to learn a robust and humanoid parking strategy efficiently. Wepresent preliminary results in a high-fidelity simulator to show our approach can outperforma geometric planning baseline in the robustness to environment noise and efficiency ofplanning.