Abstract Truss design is a highly constrained problem due to mechanical requirements and practical limitations related to fabrication and assembly. This study formulates truss design synthesis as a discrete Markov decision process, in which grammar-constrained actions generate feasible intermediate layouts and Monte Carlo Tree Search (MCTS) learns an optimal design policy. Previous work has shown that MCTS outperforms both metaheuristic methods, such as genetic algorithms, and alternative reinforcement learning approaches, including Q-learning and deep Q-learning. However, its computational scalability is limited by the rapid growth of admissible configurations in dense grid design domains. To address these limitations, we propose a Hierarchical MCTS (H-MCTS) framework in which staged grid refinements focus computational resources on promising regions of the domain, thereby alleviating the curse of dimensionality. Benchmark evaluations show that H-MCTS consistently improves design quality and reduces computational cost compared to single-stage MCTS. To accommodate variable design conditions, H-MCTS is further applied to on-the-fly structural adaptivity through an offline–online strategy that precomputes optimal solutions and interpolates them in real time. The effectiveness of the computational procedure is demonstrated on a bridge-like truss structure that is progressively constructed and then morphed to accommodate moving loads and localized damage.
Sedighzadeh et al. (Tue,) studied this question.