# Khronos: A Unified Approach for Spatio-Temporal Metric-Semantic SLAM in Dynamic Environments

**Authors:** Lukas Schmid, Marcus Abate, Yun Chang, Luca Carlone  
**arXiv:** [2402.13817](https://arxiv.org/abs/2402.13817)  
**Date:** 2024-02-21  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2402.13817v1/

## Key points

- Perceiving and understanding highly dynamic and changing environments is crucial for robot autonomy
- Progress has been made in dynamic SLAM approaches for accurate pose estimation, but there is a gap in dense spatio-temporal representations of the environment
- The authors introduce the Spatio-temporal Metric-semantic SLAM (SMS) problem and propose a framework to solve it efficiently
- Khronos approach unifies short-term and long-term dynamics, leading to real-time construction of dense spatio-temporal maps
- Through experiments, Khronos accurately captures 3D scene evolution and outperforms existing methods in map construction
- Validation on two robots operating in challenging environments demonstrates Khronos' effectiveness

## Summary

Perceiving and understanding highly dynamic and changing environments is essential for enabling robot autonomy in various applications. Significant progress has been made in developing dynamic SLAM approaches to accurately estimate the robot's pose, but there remains a gap in constructing dense spatio-temporal representations of the environment. This detailed understanding of the scene and its evolution over time is crucial for long-term robot autonomy, especially when operating alongside humans and other agents subject to short-term dynamics and long-term changes. To address this challenge, the authors introduce the Spatio-temporal Metric-semantic SLAM (SMS) problem and propose a framework to efficiently factorize and solve it. The proposed factorization leads to a natural organization of a spatio-temporal perception system where a fast process tracks short-term dynamics within an active temporal window, while a slower process reasons over long-term changes using a factor graph formulation. The implementation of this approach, named Khronos, unifies interpretations of short-term and long-term dynamics and can construct dense spatio-temporal maps in real-time. Through simulated and real-world experiments, the authors demonstrate that Khronos accurately captures the 3D scene's evolution over time. Comparing against baselines across multiple metrics, Khronos outperforms existing methods in constructing spatio-temporal maps. Furthermore, validation on two heterogeneous robots operating in challenging large-scale environments showcases the effectiveness of Khronos in real-world scenarios. In summary, Schmid et al. 's work presents a unified approach that addresses the complexities of dynamic environments for robot autonomy through efficient spatio-temporal perception. By introducing Khronos as a solution to the SMS problem, this research contributes towards enhancing robots' capabilities to navigate and interact effectively in ever-changing surroundings with humans and other agents.