This paper presents Khronos, a unified approach for spatio-temporal metric-semantic SLAM in dynamic environments. The paper introduces the Spatio-temporal Metric-semantic SLAM (SMS) problem, which aims to build a dense metric-semantic model of the world at all times as the robot navigates the scene. The proposed framework factorizes the SMS problem based on spatio-temporal local consistency, allowing for the disentanglement of errors arising from sensing noise, state estimation errors, dynamic objects, and long-term changes in the scene. The framework integrates this insight into a spatio-temporal perception system, named Khronos, which is the first real-time metric-semantic system capable of building a spatio-temporal map of the scene. The paper evaluates the method in several simulated scenes with detailed annotations on background reconstruction, object detection, motion tracking, and change detection, as well as on multiple robotic platforms navigating highly dynamic real-world environments. The contributions include formalizing the SMS problem, proposing a novel factorization of the SMS problem, and presenting Khronos, the first spatio-temporal metric-semantic perception system. The paper also discusses related works, problem statement, fragments and factorization, and the implementation of Khronos. The experiments show that Khronos outperforms baselines across multiple metrics and is able to construct a dense spatio-temporal map in real-time. The paper also discusses the importance of semantic segmentation input, spatio-temporal map beliefs, mobile robot experiments, and computation time.This paper presents Khronos, a unified approach for spatio-temporal metric-semantic SLAM in dynamic environments. The paper introduces the Spatio-temporal Metric-semantic SLAM (SMS) problem, which aims to build a dense metric-semantic model of the world at all times as the robot navigates the scene. The proposed framework factorizes the SMS problem based on spatio-temporal local consistency, allowing for the disentanglement of errors arising from sensing noise, state estimation errors, dynamic objects, and long-term changes in the scene. The framework integrates this insight into a spatio-temporal perception system, named Khronos, which is the first real-time metric-semantic system capable of building a spatio-temporal map of the scene. The paper evaluates the method in several simulated scenes with detailed annotations on background reconstruction, object detection, motion tracking, and change detection, as well as on multiple robotic platforms navigating highly dynamic real-world environments. The contributions include formalizing the SMS problem, proposing a novel factorization of the SMS problem, and presenting Khronos, the first spatio-temporal metric-semantic perception system. The paper also discusses related works, problem statement, fragments and factorization, and the implementation of Khronos. The experiments show that Khronos outperforms baselines across multiple metrics and is able to construct a dense spatio-temporal map in real-time. The paper also discusses the importance of semantic segmentation input, spatio-temporal map beliefs, mobile robot experiments, and computation time.