Emerging automated vehicles (AV) may be able to provide advanced information about the surrounding information with video cameras, radar sensors, lidar sensors, etc. Such information will enable estimating and predicting transportation system states on mobility, energy, and emissions. In this study, a physical informed neural network is developed to perform an accurate tire-road friction estimation by utilizing those advanced vehicle sensors. A runway friction tester is employed as the ground truth. More than 15,000 GPS date points and other vehicle dynamic data points are collected during the field experiment, as a result, short convergence time and desired prediction accuracy are achieved by introducing the magic tire model and the slip-slop factor into the loss function.