Amay Saxena

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2021-248

December 1, 2021

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-248.pdf

Simultaneous localization and mapping (SLAM) is the problem of jointly estimating the state (such as pose, velocity, IMU biases etc) of a robot (localization) along with a map of the environment of the robot (mapping). The typical estimators used for landmark-based visual-inertial SLAM fall into one of two categories: batch optimization methods, and filtering methods. This paper analyzes the landmark-based SLAM problem through the lens of nonlinear optimization, and presents a framework that can be used to analyze, implement, and flexibly interpolate between a vast variety of SLAM algorithms. In particular, we demonstrate the equivalence between filtering based algorithms such as the Multi-state constraint Kalman filter (MSCKF) and optimization based algorithms like the sliding window filter. We present a re-interpretation of the MSCKF in terms of nonlinear optimization, and present a novel implementation based on it. We empirically compare the performance of sliding window filters and MSCKF on challenging image sequences, and use the proposed re-interpretation to explain the relative performance characteristics of the two classes of algorithms.

Advisors: S. Shankar Sastry

---

BibTeX citation:

@mastersthesis{Saxena:EECS-2021-248,
    Author= {Saxena, Amay},
    Title= {Simultaneous Localization and Mapping Through the Lens of Nonlinear Optimization},
    School= {EECS Department, University of California, Berkeley},
    Year= {2021},
    Month= {Dec},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-248.html},
    Number= {UCB/EECS-2021-248},
    Abstract= {Simultaneous localization and mapping (SLAM) is the problem of jointly estimating the state (such as pose, velocity, IMU biases etc) of a robot (localization) along with a map of the environment of the robot (mapping). The typical estimators used for landmark-based visual-inertial SLAM fall into one of two categories: batch optimization methods, and filtering methods. This paper analyzes the landmark-based SLAM problem through the lens of nonlinear optimization, and presents a framework that can be used to analyze, implement, and flexibly interpolate between a vast variety of SLAM algorithms. In particular, we demonstrate the equivalence between filtering based algorithms such as the Multi-state constraint Kalman filter (MSCKF) and optimization based algorithms like the sliding window filter. We present a re-interpretation of the MSCKF in terms of nonlinear optimization, and present a novel implementation based on it. We empirically compare the performance of sliding window filters and MSCKF on challenging image sequences, and use the proposed re-interpretation to explain the relative performance characteristics of the two classes of algorithms.},
}

EndNote citation:

%0 Thesis
%A Saxena, Amay 
%T Simultaneous Localization and Mapping Through the Lens of Nonlinear Optimization
%I EECS Department, University of California, Berkeley
%D 2021
%8 December 1
%@ UCB/EECS-2021-248
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-248.html
%F Saxena:EECS-2021-248