An IMU is one of many sensors which can be used to estimate robot motion through integration at high data frequency (dead reckoning) while external pose corrections are not available (via GPS, landmarks, etc).

Specifically, if your IMU contains an accelerometer and a gyro you may use it to estimate linear displacement (by integrating accelerations twice in all three axes) and angular displacement (by integrating angular velocities once in all three axes).

As an added bonus, if your IMU also contains a magnetometer, you may use it to correct the pose estimate obtained by the process mentioned earlier, if you make some considerations about your own acceleration and the local gravitational and magnetic fields.

The reason why you need external pose corrections is that these sensors contain errors that, when integrated, generate a drift in your estimate (i.e. your estimate eventually diverges because of random fluctuations around the mean). So external pose corrections, even if not always available, allow you to bound the pose uncertainty.