Abstract: This paper addresses three key challenges faced by coal mine roadway cleaning robots operating in long straight underground roadways without satellite signals: localization degradation, difficulty in multi-process coordination, and labor-intensive attachment replacement. An integrated technical framework combining autonomous navigation, hierarchical collaborative control, and automatic attachment replacement is proposed. For autonomous navigation, an angle-clustering ground segmentation method is combined with a tightly coupled unscented Kalman filter (UKF) for fusion positioning, effectively suppressing axial drift. Furthermore, an attitudeaware improved A* global planning algorithm is developed to eliminate the need for terminal orientation adjustment. For collaborative control, a hierarchical finite state machine based architecture is constructed to enable autonomous multi-process scheduling. For attachment replacement, a distance-triggered coarsetofine segmented heterogeneous perception fusion strategy is introduced. This strategy continuously weights and fuses joint state feedback with visual positioning information, achieving full-process automatic attachment replacement. Underground industrial trials, with a cumulative operation of 160 hours and a total footage of 60 meters, show that the average trajectory error is 0.01 m, the terminal heading deviation is 3.8°, and the dynamic obstacle avoidance success rate is 92%. The overall process success rate reaches 80%, and the manipulator end-effector tracking error is 0.12 m. Fourteen out of fifteen?attachment replacements were successful, with an average cycle time of 4.8 min per replacement, representing a 60% efficiency improvement over manual-assisted methods. The number of operators per shift is reduced from 5 to 2 (a 60% reduction), while per-capita efficiency increases from 1.8 m3/shift to 2.6 m3/shift. The proposed framework effectively resolves the three core challenges-positioning degradation, multi-process coordination, and automatic attachment replacement-for roadway cleaning robots in unstructured confined underground environments, significantly enhancing both operational safety and overall efficiency.