In this article, we present a collection of detailed mathematical formulations and graph structures for estimating egomotion and tracking dynamic objects in Simultaneous Localization and Mapmaking (SLAM) problems. We analyze, evaluate, and test each formula using real-world datasets to identify the most accurate approach. Our contributions include introducing a front-end agnostic framework using GTSAM [1] that can be used to evaluate various Dynamic SLAM formulations.
To demystify complex concepts, let’s consider SLAM as a challenging game of "motion and mapping" where objects move, and we need to keep track of their positions and relationships while building an accurate map of the environment. Our article provides a toolkit of mathematical formulas and graph structures that help us solve this game more effectively, resulting in more stable optimization processes and the most accurate estimation in our experiments.
We break down the contributions of the paper into three main points:
- Introducing a range of detailed mathematical formulations for estimating egomotion and tracking dynamic objects in SLAM problems. These formulations provide new ways to approach the problem, offering improved accuracy and stability compared to existing methods.
- Rigorously analyzing, evaluating, and testing each formula using real-world datasets to determine which approaches are most effective. This helps us understand which formulations work best in different scenarios and why.
- Providing a front-end agnostic framework using GTSAM [1] that can be used to evaluate various Dynamic SLAM formulations. This framework allows us to compare different approaches side by side, providing valuable insights into their strengths and weaknesses.
In summary, our article offers a comprehensive toolkit of mathematical formulations and graph structures for solving the challenging problem of SLAM. By introducing new formulations and evaluating them using real-world datasets, we provide a more accurate and stable optimization process, making it easier to build accurate maps of dynamic environments.
