Learn to rotate: Part orientation for reducing support volume via generalizable reinforcement learning

In design for additive manufacturing, an essential task is to determine the optimal build orientation of a part according one or multiple factors. Heuristic search used by most methods select from large solution space. Search algorithms occasionally converge towards local optimum and waste considerable time on trial error. The above issues could be addressed if there was intelligent agent that knew search/rotation path given 3D model. A straightforward method construct such reinforcement learning (RL). By adopting this idea, time-consuming online searches in existing will moved offline stage, potentially improving performance. This challenging research problem because goal capable rotating arbitrary models, whereas RL agents frequently struggle generalize new scenarios. Therefore, paper suggests generalizable (GRL) framework train agent, GPU-accelerated GRL benchmark support training, testing, comparison approaches. Experimental results demonstrate proposed average outperforms others terms effectiveness efficiency. It proved have potential solve minima problems raised approaches, swiftly discover global (sub-)optimal (i.e. 2.62x 229.00x faster than random algorithm), beyond environment which it trained.