With acquisition protocols such as high angular resolution diffusion imaging, head motion can become an issue. Although the misalignment between diffusion-weighted images (DWIs) can be corrected in a post-processing step, this might increase partial volume effects, because of the relatively low spatial resolution of DWIs and interpolation in the registration procedure. If able to detect motion online, the scanner technician could be issued a warning and make a decision accordingly. Orientation distribution functions (ODF) can be reconstructed online using a Kalman filter (KF). We present three contributions related to the problem of online ODF reconstruction and motion detection in HARDI. First, we develop a proper error propagation accounting for the non-linear transform on the diffusion signal. Next, we develop two motion detection algorithms, based on the monitoring of residuals, and compare them using synthetic data.