For a new tiltrotor cross-domain robot,a non-parametric system identification algorithm was used to realize the dynamic modeling for the cross-domain robot with less data.By using the algorithm,based on multioutput Gaussian processes and in the absence of prior knowledge of the robot dynamic model,the robot model could be obtained by collecting experimental data.In order to reduce the computational complexity of the Gaussian process regression,local Gaussian process regression was used to identify the dynamic model of the robot.To verify the effectiveness of the identification method,the quad-tiltrotors rotor model was used as the test platform with the robot data obtained through remote control.Under the same control input of the robot,the effectiveness of the robot dynamics identification method and the rapidity of the local Gaussian process regression were proved.