Learning to Think and Explore in 3D World

Researchers are studying how vision-language models can learn to think and explore in 3D space by interacting with a 3D environment through actions and observations, using reinforcement learning driven by final-answer correctness.