This paper proposes an unscented transformation for a FastSLAM framework. The unscented transformation is used to estimate robot poses in conjunction with generic particle filter used in standard FastSLAM framework. This method can estimate robot poses more consistently and accurately than the use of single standard particle filters, especially when involving highly nonlinear models or non-Gaussian noises. In addition, our algorithm avoids the calculation of the Jacobian for motion model which could be extremely difficult for high order systems. We proposed two different sampling strategies known as a symmetrical and a spherical simplex to unscented transformation to estimate robot poses in FastSLAM framework. Simulation results are shown to validate the performance goals.