A distributed reinforcement learning control architecture for multi-link robots - experimental validation