Back to search

Rochester Institute of Technology

Adaptive Parameter Tuning for Robust Event-Based SLAM in Diverse Scenarios

Abstract

dc:description.abstract

In the ever-changing field of contemporary robotics, Visual Simultaneous Localization and Mapping (V-SLAM) stands as a crucial technology, facilitating navigation and mapping in various environments. While traditional V-SLAM approaches excel in static environments, the need for their adaptation to handle dynamic and realistic scenes is becoming increasingly apparent. This thesis delves into the advancements within the V-SLAM field, with a focus on enhancing its adaptability and performance in dynamic and challenging conditions. Initially, a comprehensive review of V-SLAM methods in the literature is presented in this thesis, offering an in-depth examination of current cutting-edge techniques, their strengths, weaknesses, and prospects for future exploration. Moreover, it offers an overview of frequently utilized datasets for assessing the efficacy of V-SLAM methodologies. In our work, we identify key limitations of existing methods, including issues with scalability in large scenes, susceptibility to lighting variations, high computational costs, and performance degradation in noisy environments. Secondly, this thesis explores integrating bio-inspired event cameras to enhance SLAM system state estimation accuracy, especially in dynamic and low-light conditions. Despite the increasing amount of research on event-based sensing, a significant deficiency exists in the literature concerning the optimization of Event SLAM parameters. To tackle this issue, we present VIO-Gradient-based Optimization (VIO-GO), an innovative framework built to automate the parameter adjustment procedure and pinpoint the most effective parameters for Event SLAM algorithms that visualize event data using edge images. Validation experiments conducted on the Event camera dataset demonstrate the superiority of our system, achieving a 60% improvement in the mean position error compared to fixed parameter sets. Across diverse sequences of the used dataset, our model adeptly determines the best parameters for a Visual Inertial Odometry (VIO) system that utilizes edge images for event data representation, underscoring its heightened precision and dependability in parameter optimization within difficult conditions. Moreover, VIO-GO proves scalability with increasing parameter complexity, resulting in a notable 24% error reduction using the largest parameter set (VIO-GO8) compared to the smallest parameter set (VIO-GO2). By automating the parameter tuning process in Event SLAM systems, this thesis contributes to the advancement of autonomous robotics, laying the groundwork for more resilient SLAM systems capable of operating effectively in dynamic real-world environments.

Degree

thesis:*
Name thesis:degree_name
Electrical Engineering (MS)
Level thesis:degree_level
Thesis
Discipline thesis:degree_discipline
Electrical Engineering

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Sakhrieh, Saber Muawiyah
Contributors dc:contributor
  • Jinane Mounsef

Subjects

dc:subject × 6

Identifiers

dc:identifier.*
Repository record dc:identifier
https://repository.rit.edu/theses/11792

Chain of custody

source
Harvested from
Rochester Institute of Technology
Base URL
repository.rit.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Sakhrieh, Saber Muawiyah. Adaptive Parameter Tuning for Robust Event-Based SLAM in Diverse Scenarios. Thesis thesis, https://repository.rit.edu/theses/11792