Abstract: The estimation of critical states of vehicles is critical to the implementation of feedforward control and the safety of the driving vehicle. Limited by the computational power of onboard equipment and the performance of sensors, the model-driven method is generally adopted to estimate states of difficult measurements, and the estimation is carried out based on simplified vehicle dynamic models. We propose a new data-driven method called Parallel Estimation Method. Combining parallel systems and deep neural networks, our method is capable to perform an online learning and to update dynamically, which makes the method more robust and have high data efficiency. Experimental results show that the proposed method has good precision and can be applied to various working conditions.