Spatially-Distributed Interactive Behaviour Generation for Architecture-Scale Systems Based on Reinforcement Learning