Using Reinforcement Learning to Optimize Core and Atria Placement

A preliminary research project has described the importance of locating the core in a high rise tower using brute force to exhaustively find all possible permutations, as well as the two evaluation methods: visibility and distance to the closest source of natural daylight. Going forward, a more complex scenario for a core placement problem is a horizontally expanded building, where cores must be deployed for fire safety regulations, flight path and egress. But not only cores are important for deep floorplates, also atria to infiltrate the building with natural daylight. Flight distance, visibility and distance to the closest source of natural daylight are quite demanding parameters for placement of cores and atria. Reinforcement Learning, where each core and atria is an agent can help to tackle this multi dimensional problem. More specifically q-learning is utilised to place cores and atria, furthermore enabling them to communicate with each other, coming up with solutions a human design team would have to work on for weeks.