Generative Adversarial Networks for Design Exploration and Refinement (GANDER)

Penn State developed methods to use artificial intelligence to explore design spaces for complex systems. The methods used game engines to model physics, and a variety of AI architectures (recurrent neural networks, generative adversarial networks) to learn the rules for generating satisfactory designs. The AI learned how to generate both physical configurations and behaviors. The methods were generated on a variety of examples, to include air vehicles, rotorcraft, soaring aircraft, and sailing vessels. Designs were analyzed computationally, and also fabricated and tested via scale models.