Research on Multi-Sensor Fusion Positioning Technology for Industrial Robots

Dynamic obstacles and structural similarities in industrial environments can easily lead to the degradation and drift of matching for single sensor positioning. To this end, this paper constructs a heterogeneous perception network of liDAR, vision, IMU and QR code, and proposes a hierarchical tightly coupled fusion architecture based on factor graphs. Compensate for point cloud distortion through IMU pre-integration and enhance the registration stability of degraded scenarios with intensity characteristics. Cross-modal keyframe association resolves the uncertainty of visual depth, and QR codes provide absolute poses to suppress global drift. Structured warehouse experiments show that the system′s errors in the X and Y directions are ±0.068 m and ±0.069 m respectively, and the heading angle error is ±0.107 rad. Compared with the single laser scheme, the accuracy is improved by 52.1%.