Towards Multi-Agent Interactive Reinforcement Learning for Opportunistic Software Composition in Ambient Environments