Zhipeng Du

To model and understand dynamic scenes is crucial in numerous vision applications for robotics and autonomous driving. In these domains, accurately reasoning over physical information such as object properties, motion, interaction, and environmental forces is essential for predicting future states and making informed decisions. Our aim is to explore the intersection of physical reasoning and simulation, enabling machine systems to simulate, understand and predict real-world outcomes in complex, dynamic environments. This project will enhance the robustness of vision models in understanding and interacting with their surroundings, ultimately leading to more reliable and efficient solutions in autonomous systems.