Designing Architecture in a Bridge Engineering Simulation
Abstract
In this paper, we used a bridge engineering simulator as an environment to search for low-cost architecture designs while enforcing close-to-real-world physical constraints on the design. The environment's built-in architecture validation functions are used in each step of the reinforcement learning episode. We designed the observation space and reward functions to enhance the bridge engineering simulator's capabilities for reinforcement learning tasks. We evaluated the performance of the proximal policy optimization (PPO) and PPO-LSTM agents using a variation of three different observation space types and two reward functions. The results show that the novel adjacency-map-based observation space and a positive step reward (in an otherwise negative reward function) produce the best performance for the PPO agent.
Department(s)
Computer Science
Document Type
Conference Proceeding
DOI
10.1109/ICPS59941.2024.10639984
Keywords
architecture, engineering, reinforcement learning, RL environment design, simulation
Publication Date
1-1-2024
Recommended Citation
Hartzler, Nathan and Belkhouche, Mohammed Y., "Designing Architecture in a Bridge Engineering Simulation" (2024). Faculty Scholarship. 424.
https://bearworks.missouristate.edu/articles00/424
Journal Title
2024 IEEE 7th International Conference on Industrial Cyber Physical Systems Icps 2024
