This paper presents a method for planning near-optimal collision-free path between two pre-specified points with given orientations for nonholonomic flying robots or UAVs navigating in a 3D space amid obstacles. The proposed method is based on the Fast Marching Method (FMM), which uses a first order numerical approximation of the Eikonal equation. The FMM is combined with average filter, B-spline functions and a new tool called Virtual Obstacle to guarantee smoothness of the produced path. The Virtual Obstacle is in the form of a torus to adjust the path with orientation of the flying robot at the start and goal points. Experimental results showed that the proposed method can find near-optimal paths with maximum curvatures less than the robot's minimum flying radius.