Normal force and particle speed are the two most important parameters in determining the mass removal rate on the workpiece in the stream finishing process. However, quantifying them is still a big challenge due to the complexity of the interaction between media flow and the workpiece. In this study, a continuum-based (μ(I) rheology) numerical model was developed to simulate the granular media flow in a robotic stream finishing (RSF) system and validated with in-house experimental data. A novel frictional wall-slip model was developed to describe the media flow speed on the surface of a rectangular workpiece. The impact of three key process variables: (i) immersion depth (D), (ii) angle of attack (θ), and (iii) radial distance (r) on the resulting normal force and media flow speed was analyzed. The resulting normal force on the workpiece strongly depends on the immersion depth. A small increase in the hydrostatic pressure is amplified through elevated effective viscosity near the workpiece, resulting in a significant increase in the normal force. A critical θ for the dependency of normal force was identified. Below the critical θ, the normal force varies slightly with the increase in θ. Above the critical θ, the normal force decreases rapidly with the increase in θ. The normal force also increases linearly with the increase in radial distance. In general, the media flow speed increases rapidly with θ. However, the increase peaks at θ = 75° due to the wake effect behind the leading edge. As D increases, the media flow speed only changes marginally, given the same θ. The media flow speed increases with radial distance, and the relationship is quadratic. We have validated – through experiments – that the numerical model can reliably predict both normal force and media flow speed on the rectangular workpiece. As the local material removal rate has a direct correlation to the normal force and the media flow speed, this numerical model is capable of predicting material removal distribution on an actual polished component and, therefore, will be an indispensable tool for conducting in-process optimization and toolpath planning.