Learning-based Adaptive Parameter Selection Framework for Navigation in Mobile Robots
Year: 2026
Type: conference-paper
Abstract: Navigation in autonomous mobile robots depends on complex parameter tuning that often lacks generalizability. This study proposes a context-aware, learning-based adaptive parameter selection framework for local planners. Focusing on a differential-drive mobile robot simulated with ROS 2 Humble and Gazebo, the method enables dynamic switching between predefined configurations (time-optimal, balanced, and safety-optimal) during navigation. A Random Forest model was trained using features such as obstacle density, proximity, and goal distance. Results demonstrate that the proposed approach outperforms fixed configurations, achieving a 97.5% success rate and an average task completion time of 18.0 s while maintaining low computational overhead.
Authors: Alperen Keser, Pınar Oğuz Ekim
Institution: İzmir University of Economics
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7202209580