Abstarct: There is always a mismatch between the nominal kinematic model embedded in the robot controller and the actual physical model due to construction errors, assembly errors, joint clearance, flexibility in the lixnks, environmental changes, and other factors. In robot control, due to the structure of the lixnks and joints, there may be some friction and joint clearance, resulting in wear and a decrease in the accuracy of the control process. Additionally, calculating the kinematics of parallel robots, especially direct kinematics, is complex and difficult. Therefore, artificial neural networks can be used in robot modeling and control. Artificial neural networks can represent non-linear relationships that occur between input and output data. Their ability to learn from examples makes them a good option for creating the necessary maps between Cartesian and joint spaces required for direct and inverse kinematics problems. Artificial neural networks are also very useful for dealing with real-time adaptive control problems, where their computation time is shorter. In this study, a neural network was used to achieve a real model of a delta robot and then control its path. Initially, the required data was extracted, where the input of the system was the joint angles of the robot and the output was the position of the end effector. By measuring the input and output values experimentally, the required data was generated. This process resulted in obtaining the real model of the robot, which was then used in the trajectory control process. The control process was also applied to the robot experimentally. The robot followed the defined trajectories with an average accuracy of 2 millimeters, and the results were validated using distance sensors.