Object-oriented autonomous navigation system for mobile robots

Abstract

Autonomous navigation is a key capability for mobile robots, enabling exploration and targetbased movement. This dissertation integrates SLAM, semantic segmentation, and the Nav2 framework in ROS2 to enhance autonomous exploration and navigation. The system employs a frontier-based exploration strategy for mapping while using SegFormer, a transformer-based model, for human detection. A depth camera estimates the 3D position of detected objects, enabling navigation. The implementation is tested in a Gazebo simulation environment to evaluate perception, mapping, and navigation performance. Nav2’s SMAC 2D planner and MPPI controller parameters are tuned to improve trajectory quality and obstacle avoidance. Experimental results indicate a balance between exploration and target-driven navigation under simulation conditions. Future work includes real-world deployment, better adaptability to dynamic obstacles, and refining the semantic segmentation model. Multi-robot collaboration and automated parameter optimization may further improve efficiency and scalability.