This paper proposes a thermopile and magnetometer-assisted INS/GPS navigation as a robust navigation method for attitude estimation compared to the conventional loosely-coupled INS/GPS method. This system aims to maintain the accuracy of attitude estimation even when autonomous UAV loses GPS signals temporarily, and to continue the flight without crashing. Accuracy of the attitude estimation is maintained by directly correcting the attitude estimates (roll, pitch, and heading) from INS by using the infrared intensity and magnetic-based sensor models. We performed the simulation to compare the accuracy of the proposed method and loosely-coupled INS/GPS. Results showed that when the GPS signal is active, the estimation accuracy of the two methods for attitude, velocity, and position are identical. Moreover, when the GPS signal is lost, all the estimates deteriorate in the conventional method, while our system maintains the accuracy of the attitude estimation and suppresses the decrease in the accuracy of the velocity and position estimates.