In Robot-Assisted Multidirectional Additive Manufacturing (RAM-AM), the travel paths (non-printing transitions) between separate printing segments are rarely planned explicitly. This often results in discontinuities, inefficient connections, and severe collision risks, particularly in moving-bed configurations where the entire part moves relative to the nozzle. We propose a decoupled planning pipeline that first computes a safe, smooth geometric path in the platform frame and subsequently generates an executable joint-space trajectory under kinematic limits. The geometric planning utilizes an A* search on an OctoMap-based voxel grid with an asymmetric layered cost map to actively steer the end-effector away from the printed structure. This raw path is smoothed using sparse B-splines, and orientations are optimized by blending Spherical Linear Interpolation (Slerp) with Artificial Potential Fields (APF) to maximize clearance. For trajectory generation, we resolve kinematic redundancy by sampling pose tolerances and selecting a globally consistent joint sequence via second-order Dynamic Programming (DP), explicitly penalizing joint jerks and singularities. Validated on a 6-DoF KUKA KR6 R900 sixx, the method is demonstrated in scenarios involving complex obstacles like holes. Results show that our approach guarantees safe obstacle avoidance where standard baselines fail, yields significantly smoother motions, and that exploiting the redundant rotation about the extrusion axis drastically reduces the inverse kinematics search effort.