Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
126 tokens/sec
GPT-4o
47 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Kimera2: Robust and Accurate Metric-Semantic SLAM in the Real World (2401.06323v1)

Published 12 Jan 2024 in cs.RO

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.

Citations (2)

Summary

  • The paper presents Kimera2, which enhances SLAM by integrating improved visual-inertial odometry, robust pose graph optimization, and versatile semantic mapping.
  • It employs advanced techniques like optimal keyframe selection, Graduated Non-Convexity for outlier rejection, and support for multiple sensor inputs to boost accuracy.
  • Ablation studies confirm that Kimera2 outperforms state-of-the-art systems such as Vins-Fusion and ORB-SLAM3, emphasizing its practical impact in diverse real-world environments.

Kimera2: Enhancements and Evaluations

Introduction to Kimera2

The original Kimera is a well-regarded open-source tool that assists in Simultaneous Localization and Mapping (SLAM) by integrating metric and semantic information through visual and inertial sensors. Since its debut, Kimera has been implemented in numerous projects spanning academia and industry. Its appeal lies in its capability to perform in real-time while offering precise and robust state estimation and mapping functionalities. The authors have now upgraded Kimera's feature set to bolster Visual Inertial Odometry (VIO) performance, enhance robustness in pose graph optimization, and improve semantic mapping capabilities.

Systematic Improvements

The improvements in Kimera2 are extensive and target both the frontend and backend processes. The frontend improvements encompass better feature tracking and more effective keyframe selection, which plays a crucial role in when and how the backend optimizes and updates the robot's trajectory and map. Kimera-VIO’s frontend, responsible for initial data processing, now accepts various forms of input such as monocular, stereo, RGB-D images, and accommodates external odometry sources like wheel or LIDAR odometry. These advancements allow for a more flexible adaptation to different robotic platforms and environments. The backend updates integrate Graduated-Non-Convexity (GNC) for outlier rejection in pose-graph optimization—an advancement that outperforms previous methods and enhances the robustness and accuracy of the entire SLAM process.

Ablation Studies and Comparisons

To quantify each added feature's impact, the authors conducted detailed ablation studies using an array of simulated and real-world datasets. For instance, adding external wheel odometry improved accuracy in outdoor environments, while image feature binning— selectively ignoring parts of an image—enhanced performance in scenarios with obstructed camera views. Moreover, refining keyframe selection led to more efficient processing without compromising the richness of the trajectory information. The paper also compares Kimera2 with other state-of-the-art SLAM systems, such as Vins-Fusion and ORB-SLAM3, across various platforms, demonstrating how Kimera2 frequently yielded superior performance, especially in large-scale datasets.

Conclusion and Future Development

The authors have achieved notable enhancements with Kimera2, advancing the original Kimera by offering an even more versatile and reliable tool for metric-semantic SLAM. Empirical results highlight its improved performance and robustness across diverse scenarios. These strides provide a solid foundation for future research and development, potentially leading to even more sophisticated and resilient spatial perception models for robotics. The open-source nature of Kimera ensures that these contributions will benefit a broad user base, propelling further innovations within the community.

Github Logo Streamline Icon: https://streamlinehq.com
X Twitter Logo Streamline Icon: https://streamlinehq.com
Youtube Logo Streamline Icon: https://streamlinehq.com