Aiming at the temperature drift phenomenon of strain gauge torque sensor, the temperature compensation model based on piecewise linear interpolation method was established, and the two-dimensional calibration data of torque and temperature were embedded into the data fusion software algorithm. According to the accuracy requirement, the data was divided into multiple different intervals and the calibration points in the intervals were replaced into the linear expression. Then the predicted torque value was output. In this way, the temperature drift of the torque sensor was corrected in real time, so that the sensor has the function of temperature self-compensation. The results show that after compensation by piecewise linear interpolation, the sensitivity temperature coefficient of the sensor is increased from to , and the relative value of temperature additional error is increased from 2.67% to 0.5%, both of which are increased by more than 5 times. The method is simple and effective. The real-time temperature compensation of torque sensor has been realized, and the temperature stability of torque sensor has been improved.