Based on the current situation that most fire-fighting robots are operated by humans and do not have independent planning and operation abilities, in this paper an intelligent fire-fighting robot is designed using multi-sensor fusion. The robot has the functions of automatic inspection and fire-fighting, and can integrate the information of the operational environment and make decisions based multi-sensor fusion. An improved path-planning mechanism is proposed in order to overcome some disadvantages of the ant colony optimization algorithm, such as its easy tendency to reach local optimal solutions, slow convergence speed and weak global searching ability. A comprehensive evaluation method of the improved ACO is established to quantify its relevance and effectiveness. A joint calibration scheme for the color and temperature information obtained using an infrared thermal imager and a binocular vision camera was designed, and the internal and external parameters and distortion coefficient of the camera were successfully obtained. Based on the principle of binocular vision, a fire source detection and location strategy is proposed. When a fire source is detected, the location of the fire source is determined quickly and rescue path planning can be carried out, which improves the intelligence level of the fire-fighting robot. Finally, MATLAB and ROS are used to analyze the improved algorithm, and a fire site patrolling experiment is carried out. The results showed that the improved ACO greatly improves the convergence, reduces the number of iterations and greatly shortens the length of the patrol path, while the robot can effectively determine the location of the fire source efficiently during independent patrols and sound alarms, which will save precious time for fire-fighting and emergency rescue personnel.