Classical molecular dynamics simulations have been performed to study the details of collision-induced desorption (CID) of nitrogen molecules adsorbed at low coverages on Ru(001). Semiempirical potential energy surfaces (PES) were used to describe the movable two layers of 56 ruthenium metal atoms each, the nitrogen adsorbate, the Ar and Kr colliders, and the interactions between them. An experimentally measured threshold energy for the CID process of 0.5 eV and the dependence of the cross section sigma(des) on incidence energy and angle of incidence have been precisely reproduced in the energy range of 0.5-2.5 eV. Strong enhancement of the sigma(des) is predicted as the angle of incidence increases. Kinetic energy and angular distributions of the scattered rare gas and the desorbing nitrogen were determined as a function of the dynamical variables of the collider. It is predicted that half of the collision energy is transferred to the solid and the other half is shared among the two scattered species. While no vibrational excitation is observed, efficient rotational energy excitation is predicted which depends on both incident energy and angle of incidence. Polar and azimuthal angular distributions were found to be strongly dependent on the incidence angle and energy of the colliders. These results suggest a new CID mechanism for the weakly chemisorbed nitrogen molecules on Ru(001), based on extensive analysis of individual trajectories. According to this mechanism, the CID event is driven by an impact excitation of frustrated rotation or tilt motion of the adsorbed molecule as a result of collision with the energetic rare gas atom. In addition, lateral motion along the surface is also excited. Strong coupling of these two modes with the motion in the direction normal and away from the surface eventually leads to desorption and completes the CID process. The efficiency of this coupling is dictated by the details of the corrugation of the Ru-N-2 PES. It is concluded that the simple hard cube-hard sphere model, frequently used to analyze CID processes, is insufficient for the description of this system. While reasonably well predicting threshold energy, it cannot explain the full dynamical picture of the CID event.