Chen Geng develops physically-grounded neural simulators of the 4D physical world by combining scalable generative modeling algorithms with fundamental principles from classical physics. His research discovers novel physical structures directly from diverse multimodal data, and uses the discovered physics to build 4D world models that are scalable, generalizable, and physically consistent. He envisions these models as high-fidelity simulators that narrow the sim-to-real gap for robot learning, and as engines for scientific discovery that infer physical laws from observation.