Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Kimera2: Robust and Accurate Metric-Semantic SLAM in the Real World

Marcus Abate, Yun Chang, Nathan Hughes, Luca Carlone

TL;DR

Kimera2 advances robust, real-world metric-semantic VI-SLAM by upgrading Kimera's VI-SLAM stack (Kimera-VIO) to support multiple sensor modalities and external odometry, and by integrating modern outlier rejection (GNC) into RPGO and PGMO. The work provides extensive ablations and cross-platform experiments, evaluating against state-of-the-art pipelines (e.g., ORB-SLAM3, Vins-Fusion) across diverse platforms including drones, quadrupeds, wheeled robots, and simulated cars, demonstrating broad improvements in localization and dense mesh quality. Key contributions include feature binning, disparity-based keyframe triggering, and joint pose-graph/dense-mesh optimization via PGMO, all released open-source for community use. Overall, Kimera2 shows strong, versatile performance in large-scale and challenging environments, reinforcing its role as a practical, extensible VI-SLAM and semantic mapping solution.

Abstract

We present improvements to Kimera, an open-source metric-semantic visual-inertial SLAM library. In particular, we enhance Kimera-VIO, the visual-inertial odometry pipeline powering Kimera, to support better feature tracking, more efficient keyframe selection, and various input modalities (eg monocular, stereo, and RGB-D images, as well as wheel odometry). Additionally, Kimera-RPGO and Kimera-PGMO, Kimera's pose-graph optimization backends, are updated to support modern outlier rejection methods - specifically, Graduated-Non-Convexity - for improved robustness to spurious loop closures. These new features are evaluated extensively on a variety of simulated and real robotic platforms, including drones, quadrupeds, wheeled robots, and simulated self-driving cars. We present comparisons against several state-of-the-art visual-inertial SLAM pipelines and discuss strengths and weaknesses of the new release of Kimera. The newly added features have been released open-source at https://github.com/MIT-SPARK/Kimera.

Kimera2: Robust and Accurate Metric-Semantic SLAM in the Real World

TL;DR

Kimera2 advances robust, real-world metric-semantic VI-SLAM by upgrading Kimera's VI-SLAM stack (Kimera-VIO) to support multiple sensor modalities and external odometry, and by integrating modern outlier rejection (GNC) into RPGO and PGMO. The work provides extensive ablations and cross-platform experiments, evaluating against state-of-the-art pipelines (e.g., ORB-SLAM3, Vins-Fusion) across diverse platforms including drones, quadrupeds, wheeled robots, and simulated cars, demonstrating broad improvements in localization and dense mesh quality. Key contributions include feature binning, disparity-based keyframe triggering, and joint pose-graph/dense-mesh optimization via PGMO, all released open-source for community use. Overall, Kimera2 shows strong, versatile performance in large-scale and challenging environments, reinforcing its role as a practical, extensible VI-SLAM and semantic mapping solution.

Abstract

We present improvements to Kimera, an open-source metric-semantic visual-inertial SLAM library. In particular, we enhance Kimera-VIO, the visual-inertial odometry pipeline powering Kimera, to support better feature tracking, more efficient keyframe selection, and various input modalities (eg monocular, stereo, and RGB-D images, as well as wheel odometry). Additionally, Kimera-RPGO and Kimera-PGMO, Kimera's pose-graph optimization backends, are updated to support modern outlier rejection methods - specifically, Graduated-Non-Convexity - for improved robustness to spurious loop closures. These new features are evaluated extensively on a variety of simulated and real robotic platforms, including drones, quadrupeds, wheeled robots, and simulated self-driving cars. We present comparisons against several state-of-the-art visual-inertial SLAM pipelines and discuss strengths and weaknesses of the new release of Kimera. The newly added features have been released open-source at https://github.com/MIT-SPARK/Kimera.
Paper Structure (14 sections, 1 figure, 7 tables)

This paper contains 14 sections, 1 figure, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Kimera2: System Improvements
  4. Kimera-VIO Frontend
  5. GNC for Kimera-RPGO and Kimera-PGMO
  6. Experiments
  7. Datasets
  8. External Odometry
  9. Feature Binning
  10. Keyframe Logic
  11. GNC vs PCM
  12. PGMO
  13. Competitor Comparison
  14. Conclusions

Figures (1)

  • Figure 1: Overview of some of the platforms and datasets used in the experimental evaluation of Kimera. (a) Clearpath Jackal Robot (left), Unitree A1 quadruped (right). (b) Handheld Jetson-based sensor rig, not discussed in this paper, but evaluated in Rosinol21ijrr-Kimera. (c) uHumans2 simulator office scene. (d) CarSim simulator scene. (e) Self-driving car, not discussed in this paper, but evaluated in Abate23iser-KimeraSelfDriving.