Abstract: The high-risk firefighting environment poses severe challenges to the autonomy and efficiency of firefighting equipment. To address the common issues of existing firefighting robots, namely their reliance on external water supplies, insufficient capability to traverse complex terrains, and limited intelligence, we develop an intelligent firefighting robot integrating a large-capacity self-contained extinguishing agent, a high-mobility chassis, and advanced autonomous intelligence. The robot adopts a tracked mobile platform equipped with a self-contained agent system and a hydraulic push shovel assistance mechanism, and incorporates a tightly coupled fusion perception system centered on 4D millimeter-wave radar and LiDAR, thereby resolving the problems of robust localization and precise motion control in dense smoke and unstructured terrains. Furthermore, a dual-branch lightweight fire source recognition network that fuses RGB and thermal imaging features is proposed, enabling high-precision dynamic fire source identification and water-efficient fire suppression. Systematic comparative experiments demonstrate that in simulated complex fire scenarios, the robot achieves a mean average precision (mAP@0.5) of 98.7% for fire source recognition, representing a 12.5% improvement over a single RGB model. In actual oil-pool fire suppression tests, the single-operation success rate reaches 100%, the average fire suppression time is reduced by approximately 54% compared to traditional remote-controlled robots, and water consumption per unit area is reduced by 40%. The research findings provide a practically viable innovative solution for unmanned firefighting in high-risk environments.