Institutional Repository of Vilnius University
Sustiprintojo mokymosi taikymas LiDAR pagrįstai TurtleBot3 navigacijai trajektorijoja nužymėta vartais /
Abstract
dc:descriptionThis bachelor thesis investigates LiDAR-based navigation of a TurtleBot3 Burger robot along a trajectory marked by gates. Each gate is represented by two vertical poles, and the robot has to pass three gates in the correct order without colliding, missing a gate, or exceeding the episode time limit. The final controllers are evaluated in Gazebo through ROS 2 topics. A separate two-dimensional Gymnasium environment is used for training, model development, and reward tuning, but it is not used as the final source of quantitative evaluation. The work compares three non-privileged methods. The first method is a simple gate-middle tracking controller that extracts compact LiDAR clusters, selects a plausible gate or free-gap direction, and converts the target angle into linear and angular velocity commands. The second method is a PPO policy trained on 360 normalized LiDAR measurements and odometry-based motion features. The third method combines a LiDAR geometry helper module with PPO: the helper estimates a base action from the scan, and the policy learns a bounded residual correction. The main Gazebo evaluation consists of 100 episodes per method using the same three-gate scenario and recorded initial-condition ranges. The simple gate-middle controller achieved 95 successful episodes out of 100 and was the fastest successful method, with an average successful episode time of 17.94 s. The PPO policy with unprocessed LiDAR and odometry achieved 48 successes, with 37 collisions and 15 timeout cases. PPO with the LiDAR geometry helper also achieved 95 successes, with 2 collisions and 3 missed gates, but it was slower than the simple controller. In this scenario, the results suggest that a structured LiDAR geometry representation can improve the reliability of PPO-based control, while a hand-written geometric controller remains a strong and efficient baseline for this particular gate-following task.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Narmontas, Vytenis,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308338117&prefLang=en_US