With advancing technologies, robotic manipulators and visual environment sensors are becoming cheaper and more widespread. However, robot control can be still a limiting factor for better adaptation of these technologies. Robotic manipulators are performing very well in structured workspaces, but do not adapt well to unexpected changes, like people entering the workspace. We present a method combining 3D Camera based workspace mapping, and a predictive and reflexive robot manipulator trajectory estimation to allow more efficient and safer operation in dynamic workspaces. In experiments on a real UR5 robot our method has proven to provide shorter and smoother trajectories compared to a reactive trajectory planner in the same conditions. Furthermore, the robot has successfully avoided any contact by initialising the reflexive movement even when an obstacle got unexpectedly close to the robot. The main goal of our work is to make the operation more flexible in unstructured dynamic workspaces and not just avoid obstacles, but also adapt when performing collaborative tasks with humans in the near future.