Search and rescue (SAR) operations in post-disaster rubble environments face critical challenges from structural instability, low visibility, toxic gas accumulation, and degraded communications, while existing teleoperated unmanned ground vehicles (UGVs) suffer from high operator fatigue, high latency, and single-modal sensor failure under smoke or zero-light conditions. The problem addressed by this research is the design of an autonomous UGV capable of reliable victim detection and safe navigation in such hazardous, GPS-denied environments. A design-science and experimental engineering methodology was adopted to develop the proposed platform, SELIX, which integrates a hierarchical dual-processing architecture combining an NVIDIA Jetson Orin Nano for high-level perception and Simultaneous Localization and Mapping (SLAM) with an STM32 microcontroller for low-level real-time actuation. Data were collected from an integrated sensing suite comprising RGB and long-wave-infrared thermal cameras, a 2D LiDAR, an inertial measurement unit (IMU), ultrasonic sensors, and a gas-sensor array, together with an annotated image dataset of 2,500 samples (split 70% training, 15% validation, 15% testing) built by merging open-source data with additional site-representative images. Detection was performed using an anchor-free, TensorRT-optimized YOLOv8 model trained on fused thermal-visible imagery; localization relied on an Extended Kalman Filter (EKF) fusing wheel odometry, IMU, and LiDAR readings; and navigation combined global A* path planning with a hazard-aware Dynamic Window Approach informed by real-time gas telemetry. System performance was evaluated using precision, recall, mean Average Precision (mAP@0.5), inference latency, localization drift, and mission success rate across 50 integrated search-and-rescue trials. Results show that the thermal-visible YOLOv8 pipeline achieved 92.4% precision and 91.2% mAP@0.5 at 28 ms inference latency, improving detection success by 15% over RGB-only imagery under low-light conditions; the EKF-SLAM engine maintained 93% localization accuracy and kept trajectory drift below 0.24 m on severe rubble terrain; and the integrated platform achieved an 88% overall mission success rate with an average task completion time of 4.5 minutes. These findings position SELIX as a scalable, high-throughput solution for real-time victim localization and hazard assessment in disaster response missions.