Random drift is the main factor that influences the precision of micro-electro-mechanical system（MEMS）gyroscope.In order to real-time estimate and compensate random drift and improve on heuristic drift reduction algorithm self-adaptive drift reduction algorithm was proposed.Allan-variance analysis method was adopted to determine the null bias stability time of gyro.Random drift and true angular rate were distinguished by checking the stationary character of gyro.Test had been done with the platform of gyro-equipped indoor mobile robot Pioneer3-AT.When the body has a constant true angular rate

the new algorithm is still useful while the basic heuristic drift reduction algorithm is not use.To compare with common algorithm

the new method can increase the precision by 2 times.