Abstract: In response to the contradiction between the demand for accuracy in data collection for complex scenarios in the protection of historical towns and the insufficient adaptability of robots, this study takes the Taoyangli Historical Block in Jingdezhen as the research object and proposes a scheme for modifying quadruped robot connectors through the collaboration of AI generative algorithms and 3D printing. Based on constraints such as narrow ancient building lanes (minimum 0.8m) and fragile ground (bearing capacity ≤ 150kg/m²), through parametric modeling and multi-objective optimization algorithms, the lightweight design of the scanning bracket (699g, 53% lighter than traditional designs) and the integration of multi-modal sensors (compatible with six types of equipment) are realized. Selective Laser Melting (SLM) technology is used to manufacture the topology-optimized structure. In the empirical study of the 13.2-hectare protected area in Taoyangli, the modified robot achieved 7 hours of continuous operation, with the efficiency of multi-dimensional data collection improved by 32%, and maintained zero structural damage in the scenario of falling from 30mm-level steps. The research shows that the generative design paradigm based on scenario feature feedback can provide a highly adaptable technical path for the protection of historical towns, and the constructed "terrain parameters - sensor group - material process" trinity input model provides a reusable technical framework for the digital protection of similar cultural heritage.