Continuous-Time Fixed-Lag Smoothing for LiDAR-Inertial-Camera SLAM (2302.07456v1)

Abstract: Localization and mapping with heterogeneous multi-sensor fusion have been prevalent in recent years. To adequately fuse multi-modal sensor measurements received at different time instants and different frequencies, we estimate the continuous-time trajectory by fixed-lag smoothing within a factor-graph optimization framework. With the continuous-time formulation, we can query poses at any time instants corresponding to the sensor measurements. To bound the computation complexity of the continuous-time fixed-lag smoother, we maintain temporal and keyframe sliding windows with constant size, and probabilistically marginalize out control points of the trajectory and other states, which allows preserving prior information for future sliding-window optimization. Based on continuous-time fixed-lag smoothing, we design tightly-coupled multi-modal SLAM algorithms with a variety of sensor combinations, like the LiDAR-inertial and LiDAR-inertial-camera SLAM systems, in which online timeoffset calibration is also naturally supported. More importantly, benefiting from the marginalization and our derived analytical Jacobians for optimization, the proposed continuous-time SLAM systems can achieve real-time performance regardless of the high complexity of continuous-time formulation. The proposed multi-modal SLAM systems have been widely evaluated on three public datasets and self-collect datasets. The results demonstrate that the proposed continuous-time SLAM systems can achieve high-accuracy pose estimations and outperform existing state-of-the-art methods. To benefit the research community, we will open source our code at ~\url{https://github.com/APRIL-ZJU/clic}.

• The paper presents a continuous-time fixed-lag smoothing framework that integrates LiDAR, inertial, and camera data for enhanced SLAM performance.
• It employs a B-spline continuous-time model within a factor-graph optimization context to achieve precise temporal alignment and motion distortion correction.
• The method demonstrates robust real-time performance across diverse datasets, matching or exceeding state-of-the-art SLAM accuracy even in GPS-denied environments.

Continuous-Time Fixed-Lag Smoothing for LiDAR-Inertial-Camera SLAM

This paper presents an innovative approach to Simultaneous Localization And Mapping (SLAM) using a continuous-time fixed-lag smoothing method, focusing on LiDAR-Inertial-Camera (LIC) integration. The authors address a significant challenge in multi-modal SLAM systems—efficiently and accurately fusing asynchronous sensor data sampled at different time frequencies. By leveraging continuous-time trajectory estimation, they enable high-precision sensor fusion that leads to enhanced localization performance.

The primary contribution of this work lies in the development of a continuous-time fixed-lag smoothing framework within a factor-graph optimization context. This method allows for querying poses at any desired time instant, providing flexibility and adaptability for various sensor update rates and intervals. The authors employ a B-spline-based continuous-time model, which inherently supports accurate motion distortion correction and precise temporal alignment of sensor data. Unlike traditional discrete-time approaches, the continuous-time framework effectively addresses the dynamic interaction between different sensor modalities without necessitating complex synchronization hardware.

One of the noteworthy aspects of this method is its capacity for real-time processing, even considering the intricacies introduced by continuous-time formulations. The authors achieve this by maintaining a constant-size sliding window and applying probabilistic marginalization to effectively handle the computational demands. This approach strategically reduces the complexity associated with optimizing large-scale factor graphs, a common requirement in real-time SLAM systems.

The practical implementation of their technique is realized in several SLAM configurations, encompassing LiDAR-Inertial (LI) systems and full LiDAR-Inertial-Camera (LIC) systems, even extending to multi-LiDAR setups. This paper provides extensive evaluation across three public datasets and self-collected datasets, demonstrating that the proposed systems match or surpass state-of-the-art methods, particularly in terms of accuracy and robustness. The results indicate the approach’s efficacy in diverse environmental conditions, handling scenarios where traditional techniques might struggle, such as in GPS-denied environments or when facing high dynamic ranges in sensor data.

Although the proposed framework is computationally intensive due to its continuous-time nature and needs for iterative Jacobian computations, the authors effectively mitigate these challenges with analytical Jacobians. They also facilitate on-the-fly temporal calibration of sensor offsets, showcasing the adaptability and precision possible with continuous-time modeling.

Looking forward, the implications of this research are multifaceted. The continuous-time fixed-lag smoothing approach presents substantial opportunities for future SLAM developments, particularly in applications requiring high precision under uncertain timing conditions, such as autonomous vehicles and aerial robotics. Moreover, the ease of integrating additional sensors into the framework opens possibilities for even broader SLAM applications, potentially enhancing capabilities in areas like augmented reality, industrial automation, and personal robotics.

In conclusion, the paper introduces a sophisticated SLAM approach transcending conventional discrete-time methods, establishing continuous-time fixed-lag smoothing as a viable and effective strategy for multi-sensor fusion. The authors set a foundation for advancing SLAM accuracy and efficiency, with clear implications for ongoing research and technological innovation in autonomous navigation and related domains.