This video shows deep reinforcement learning applied to the problem of corridor coordination of MEP systems. M2x.ai used Google Tensorflow to train an AI how to coordinate the pipes, conduit and duct that traverse a corridor.
The machine created 1.2 million solutions - all buildable and without clashes - as part of the deep reinforcement process. This work was research intended to explore approaches to MEP autorouting. To our knowledge, it was the first in the world (September 2019) to use deep reinforcement learning to spatially coordinate MEP systems.