by Researchers from Purdue University’s Department of Computer Science and Institute for Digital Forestry, in collaboration with Sören Pirk from Kiel University in Germany, have made a new breakthrough in the field of artificial intelligence (AI). Their study shows that AI can now simulate the growth and shape of trees based on their environment. By drawing inspiration from the DNA molecule, the team developed AI models that compress the information needed to encode tree forms into a compact neural model.
Once trained, these AI models can generate complex tree models with detailed geometry, spanning several gigabytes. The models learn from large data sets to replicate the natural development of trees. This research is significant for various sectors, such as architecture, urban planning, gaming, and entertainment, where realistic tree designs are essential for attracting clients and engaging audiences.
Professor Bedrich Benes, along with his team, has been working with AI models for nearly a decade. Their goal was to enhance existing methods for creating digital tree models. However, the team was surprised by the models’ size despite their complexity. Benes comments, “It’s complex behavior, but it has been compressed to rather a small amount of data.”
The papers published in ACM Transactions on Graphics and IEEE Transactions on Visualizations and Computer Graphics detail the team’s research on creating the tree-simulation AI models. Co-authors Jae Joong Lee and Bosheng Li, both graduate students in computer science at Purdue University, contributed to the ACM paper. The IEEE paper included co-authors Xiaochen Zhou, another Purdue graduate student in computer science, Songlin Fei, the Dean’s Chair in Remote Sensing and director of the Institute for Digital Forestry, and Sören Pirk from Kiel University in Germany.
The researchers utilized deep learning, a branch of machine learning within AI, to generate growth models for various tree species like maple, oak, pine, and walnut, with and without leaves. Deep learning involves training AI models to perform specific tasks through interconnected neural networks that mimic human brain functionalities.
While AI has been successful in various applications unrelated to nature, such as computer-aided design and digital manufacturing, the researchers believe this breakthrough expands the capabilities of AI in modeling natural phenomena.
