We propose a hybrid approach to motion planning for redundant robots, which combines a powerful control framework with a sampling-based planner. We argue that a suitably chosen task controller already manages a huge amount of trajectory planning work. However, due to its local approach to obstacle avoidance, it may get stuck in local minima. Therefore we augment it with a globally acting planner, which operates in a lower-dimensional search space, thus circumventing the curse of dimensionality afflicting modern, many-DoF robots.