An application of the digital extended Kalman filter has been presented for the problem of estimating the absolute orientation of SemiQuad, a 2D dynamically stable walking robot. There are some differences between simulations and experiments, but the results still prove the ability of our observer to yield a short term estimation of the orientation of the robot. The precision is sufficient for the estimation to be useful for calculating the dynamic model and monitoring the balance of the robot. This task is important for SemiQuad, and vital for bipeds, to which the idea is also applicable (see Lebastard, Aoustin, & Plestan 2006 for a different type of observer). Given the possibility to re-initialize the observer at each double support phase, even if very short, as it can be for bipeds, the observer can provide the absolute orientation over time without the use of any additional sensor, which was the goal of our work.