Nowadays, autonomous robots are capable of replacing people with hard work or in dangerous environments, so this field is rapidly developing. One of the most important tasks in controlling these robots is to determine its current position. The Global Positioning System (GPS) was originally developed for military purposes but is now widely used for civilian purposes such as mapping, navigation for land vehicles, marine, etc. However, GPS has some disadvantages, like the update rate is low or sometimes the satellites’ signal is suspended. Another navigation system is the Inertial Navigation System (INS) can allow you to determine position, velocities, and attitude from the subject’s status, like acceleration and rotation rate. Essentially, INS is a dead-reckoning system, so it has a huge cumulative error. An effective method is to integrate GPS with INS, in which the center is a nonlinear estimator (e.g., the Extended Kalman filter) to determine the navigation error, from which it can update the position the object more accurately. To improve even better accuracy, this paper proposes a new method that combines the original integrated GPS/INS with tri-axis rotation angles estimation and velocity constraints. The experimental system is built on a low-cost IMU with a tri-axis gyroscope, accelerometer, and magnetometer, and a GPS module to verify the model algorithm. Experiment results have shown that the rotation angles estimator helps us to determine the Euler angles correctly, thereby increasing the quality of the position and velocity estimation. In practice, the accuracy of the roll and pitch angle is 2 degrees; the error of the yaw angle is still large. The achieved horizontal accuracy is 2m when the GPS signal is stable and 3m when the GPS signal is lost in a short period. Compared with individual GPS, the error of the integrated system is about 10% smaller.