#researchgoogle
Researchers from ETH Zürich and Google have developed a method named DIMOS for creating realistic sequences of human interactions with indoor 3D environments. Using a technique based on Markov decision processes, the method learns from motion capture data to create various human behaviors like walking around a room or sitting on furniture. The system uses a series of policies to guide the generated human figure, taking into account factors like which areas can be walked on and how close the figure is to objects. The method can adapt to different objects and scenes, allowing virtual humans to interact with a variety of indoor settings, from sitting on uniquely shaped chairs to navigating reconstructed real-world locations.