Abstract: In order to improve the position accuracy of the end effector of the 6-DOF joint robot, this paper studies the method of gravity error recognition and compensation based on the self-designed 6-DOF joint robot. Firstly, the relationship model between the angle of each joint and the position coordinates of the end of the robot is established based on D-H method, and the actual torsional stiffness of the main joint is calculated by Levenberg-Marquardt. Secondly, the static analysis of the robot under the action of gravity field under different poses is carried out using finite element analysis software, and the BP neural network model with the pose parameters as input and the end deformation as output is established to realize the online prediction of the end position error caused by the bending deformation of the connecting rod under different poses. Finally, based on the self-designed special probe, the robot end position error compensation experiment is carried out. The experimental data show that the positioning accuracy after compensation is increased by 93.6% and 92.2% in the X and Z directions, respectively.