RNA velocity techniques have emerged as efficient tools for unraveling the complex trajectories of cell development and differentiation. However, most of existing RNA velocity approaches are constrained by estimating transcriptional parameters for each gene in isolation and neglects the regulatory relationships among genes, which limits the ability to jointly resolve the dynamic rewiring of gene regulation and the underlying gene transcriptional kinetics across cell state transitions. To address these limitations, we present GRAVITY, a novel deep learning framework that explicitly integrates regulatory dynamics into transcriptional kinetics inference and utilizes a refined two-stage optimization strategy. Benchmarking across various simulated and real single-cell RNA sequencing datasets demonstrates that GRAVITY accurately infers both cellular and gene trajectories, along with their associated kinetic parameters. Most importantly, GRAVITY uncovers terminal cell states L5/L6 in embryonic brain development dataset. Furthermore, GRAVITY not only provides mechanistic insights by identifying the driver regulatory factors and modules governing cell fate, but also enables the systematic in silico simulation of cellular velocity changes induced by targeted regulatory perturbations.