This paper presents a novel perception-cognition-execution architecture for intelligent service robots, validated in financial automation through 12 invoice types. By building a three-layer system of perception layer, decision layer, and execution layer, autonomous task processing in complex business scenarios is realized. Taking the financial reimbursement scenario as a verification case, an intelligent robot system that includes 12 types of invoice processing is designed. Cross-system data capture is realized through the UiBot platform, and automated processing is supported by leveraging OCR and IDP (Intelligent Document Processing) interfaces. The system demonstrates significant potential for improving efficiency and reducing errors in financial automation tasks. The research results provide a technical solution for environmental understanding, autonomous decision-making and precise execution of intelligent service robots in dynamic environments and provide a reference for the intelligent transformation of complex business scenarios.