Abstract: High-risk fire environments demand highly autonomous and efficient firefighting equipment. To address the limitations of existing firefighting robots such as reliance on external water supply, insufficient ability to navigate complex terrain, and limited intelligence, this paper presents an intelligent firefighting robot with self-contained medium, an obstacle-crossing chassis, and advanced autonomy. The robot features a tracked design with hydraulic bulldozing assistance and a perception system based on fused 4D mmWave radar and LiDAR, enabling robust localization and motion control in smoky, unstructured environments. Meanwhile, a lightweight dual-branch network combining RGB and thermal imaging is proposed for precise, water-efficient dynamic fire source identification. Experiments show that in simulated fire scenarios, the robot achieves 98.7% mAP@0.5 in fire detection—12.5% higher than an RGB-only model. During kilometer-level autonomous navigation, its absolute trajectory error (ATE) remains below 20.5 m. In actual oil pan fire extinguishing tests, it achieved a 100% success rate, reduced extinguishing time by about 54%, and lowered water consumption per unit area by 40% compared to traditional remote-controlled robots. The research results provide an innovative solution with practical combat capabilities for unmanned firefighting in high-risk areas.