Authors: Krishna Pratap Singh Gaur, Om Prakash Sondhiya
Abstract: Innovative solutions that reduce human exposure to life-threatening risks are required because to the increasing frequency and severity of fire accidents in industrial settings, including petrochemical refineries, warehouses, power generation facilities, and chemical manufacturing plants. The methodical design, development, and experimental validation of an autonomous firefighting robot created especially for industrial deployment are presented in this work. The suggested platform combines an embedded multi-sensor array consisting of a FLIR Lepton 3.5 infrared thermal imager, Hamamatsu UV flame sensors, a Velodyne VLP-16 three-dimensional LiDAR, and electrochemical gas detectors with a thermally insulated omnidirectional Mecanum-wheel chassis. On an NVIDIA Jetson Orin NX computation module, FireDetNet-v2, a lightweight convolutional neural network trained on 45,000 annotated industrial fire pictures, achieves a mean average precision (mAP@0.5) of 97.6% at 30 frames per second. A 150-liter onboard water-AFFF suppression module uses a two-degree-of-freedom pan-tilt nozzle gimbal to administer agent at up to 12 bar, with a maximum throw range of 15 meters. GPS-denied autonomous navigation is made possible via simultaneous localization and mapping (SLAM) using the Cartographer framework using VLP-16 LiDAR data.
