Muhammad Salam Pararta Saragi, Deden Pradeka, Anugrah Adiwilaga, Muhammad Taufik Dwi Putra, Wirmanto Suteddy, Jezzy Putra Munggaran, Fauzi Abdul Rohim, Dyah Kusuma Dewi, Roni Permana Saputra
This paper presents a semantic hazard mapping system for mobile robots, designed to enhance environmental understanding by integrating object segmentation with SLAMbased mapping. The system employs a 360-degree array camera consisting of six RGB cameras, combined with a 2D ray sensor, to detect and localize hazardous objects in unstructured indoor environments. Using a custom-trained YOLO model and transfer learning, it segments objects and assigns hazard levels, which are then transformed into spherical coordinates. These semantic labels are embedded directly into the occupancy grid produced by the GMapping SLAM algorithm. Experimental results in a Gazebo simulation show consistent mapping with hazard-level annotations, achieving an average RMSE of 0.093 and a mapping cycle time of 893 ms. The method offers reliable performance for near real-time semantic SLAM. However, manual field-of-view calibration is currently required. © 2025 IEEE.
Universitas Pendidikan Indonesia, Bandung, Indonesia; Telkom University, Bandung, Indonesia; Research Center for Smart Mechatronics, National Research and Innovation Agency, Bandung, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock