Numerical simulation of physical phenomena is a powerful tool embraced by scientists and engineers for decades. Using simulation tools to predict human movements is particularly critical for training AI-enabled robots interacting with humans, providing testbeds for wearable device design, and generating a wealth of labeled, high-fidelity human motion data. However, existing physics simulators and motor control algorithms for modeling human movements were developed with a fictitious “average human” in mind, while in reality we are often more interested in predicting the motion of a particular real person. Pursuing the quest towards building personalized predictive human models, we develop a learnable and differentiable physics simulator to harness the power of data, and a data acquisition pipeline to provide large-scale biomechanical motion data for the learnable simulator to consume. These computational tools can potentially simulate a wide range of scenarios, but also provide the option to be personalized to specific individuals using only a moderate amount of data. While we focus on the application domains related to human movements, these tools are general and applicable to other robotic research problems on optimal control and parameter estimation. In this talk, I will describe our overall vision on perusing personalized predictive human models, as well as a collection of projects that advanced the state of the art toward this vision.