This paper introduces Khronos, a unified approach for spatio-temporal metric-semantic SLAM (Spatio-temporal Metric-semantic SLAM, SMS) in dynamic environments. The authors address the challenge of understanding and reasoning about short-term dynamics and long-term changes in the environment, which is crucial for robot autonomy. Khronos is designed to build dense metric-semantic representations of the environment over time, allowing robots to operate effectively in dynamic settings, such as shared spaces with humans and other agents. The key contributions of the paper include: 1. **Formalization of the SMS Problem**: The authors define the SMS problem, which aims to build a dense metric-semantic model of the environment at all times. 2. **Novel Factorization**: They propose a novel factorization of the SMS problem based on spatio-temporal local consistency, which decouples the dependencies between sensing noise, state estimation errors, dynamic objects, and long-term changes. 3. **Implementation of Khronos**: The authors present Khronos, a real-time spatio-temporal metric-semantic perception system, which includes algorithms for asynchronous local mapping and deformable global change detection. 4. **Evaluation**: The method is evaluated in simulated and real-world environments, demonstrating its ability to accurately reconstruct the scene over time and outperforming baselines across multiple metrics. The paper also discusses the limitations of the approach, such as sensitivity to partial observations and occlusions, and suggests future improvements, including the integration of modern object pose and shape estimation techniques.This paper introduces Khronos, a unified approach for spatio-temporal metric-semantic SLAM (Spatio-temporal Metric-semantic SLAM, SMS) in dynamic environments. The authors address the challenge of understanding and reasoning about short-term dynamics and long-term changes in the environment, which is crucial for robot autonomy. Khronos is designed to build dense metric-semantic representations of the environment over time, allowing robots to operate effectively in dynamic settings, such as shared spaces with humans and other agents. The key contributions of the paper include: 1. **Formalization of the SMS Problem**: The authors define the SMS problem, which aims to build a dense metric-semantic model of the environment at all times. 2. **Novel Factorization**: They propose a novel factorization of the SMS problem based on spatio-temporal local consistency, which decouples the dependencies between sensing noise, state estimation errors, dynamic objects, and long-term changes. 3. **Implementation of Khronos**: The authors present Khronos, a real-time spatio-temporal metric-semantic perception system, which includes algorithms for asynchronous local mapping and deformable global change detection. 4. **Evaluation**: The method is evaluated in simulated and real-world environments, demonstrating its ability to accurately reconstruct the scene over time and outperforming baselines across multiple metrics. The paper also discusses the limitations of the approach, such as sensitivity to partial observations and occlusions, and suggests future improvements, including the integration of modern object pose and shape estimation techniques.