Back to search

Institutional Repository of Vilnius University

Application of reinforcement learning methods for line following with an autonomous robot /

Abstract

dc:description

This thesis investigates the application of reinforcement learning (RL) methods for the line-following task using domain randomization techniques to transfer a trained policy from a simulated environment to a physical TurtleBot Burger robot. A simulation environment based on the PyBullet physics engine and the Gymnasium library was adapted for the TurtleBot Burger robot, including its physical properties, camera input, and control mechanism. Deep Q-Network (DQN), Proximal Policy Optimization (PPO), and Soft Actor-Critic (SAC) algorithms were trained and compared in the simulated environment. Although PPO achieved the highest performance in simulation, it failed to follow even a simple line in the real world. SAC was selected for sim-to-real transfer as it offered the best balance between simulated performance and real-world stability. Four domain randomization techniques were implemented and used to fine-tune the SAC model. The techniques were then tested across four different real-world tracks. The results indicate that combining all domain randomization techniques leads to, on average, better performance than applying them individually. The results also show that the techniques used in this thesis were not sufficient to produce a reliable line-following model across all tested tracks. The findings suggest that performance in a simulation environment does not necessarily correlate with performance in the real world, and that more extensive domain randomization methods are required for a reliable sim-to-real transfer.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Viliušis, Jokūbas,

Subjects

dc:subject × 1

Rights

dc:rights
Statement dc:rights
  • info:eu-repo/semantics/openAccess
Language dc:language
eng

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Vilnius University
Base URL
epublications.vu.lt/oai
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Viliušis, Jokūbas,. Application of reinforcement learning methods for line following with an autonomous robot /. Institutional Repository of Vilnius University, 2026. https://repository.vu.lt/VU:ELABAETD308339160&prefLang=en_US