Abstract: Logistics AGV robot positioning usually relies on a single sensor, leading to diminished positioning accuracy due to environmental and data precision issues, as well as safety concerns in human-machine collaboration within smart workshops. This study proposes the integration of laser radar and RGB-D cameras for mapping and positioning, further introducing AR visual tags. Additionally, a multi-sensor framework based on extended Kalman filtering is devised, leveraging multiple sensor inputs to elevate positioning accuracy to 6mm. Simultaneously, there are safety hazards present in human-machine collaboration within smart workshops. Real-time unsafe recognition using the VGG16 convolutional neural network is employed to detect potential safety risks. Moreover, the logistics AGV’s camera performs safety monitoring during the robot’s empty return journey, preventing safety incidents in smart workshops. In comparison to conventional methods, this research achieves more precise positioning and reliable environmental safety monitoring.