AutoVSR: Automatic Visual-to-Symbolic Reasoning for Symbolic Expression Generation from Circuit Schematics

AutoVSR is a framework for generating symbolic circuit expressions from schematic images. It first reconstructs a diagram into an executable intermediate representation (Executable IR), using component-rule retrieval and verification feedback, then performs multi-step derivation through a planning agent equipped with symbolic tools. The paper reports 30.01–59.45% accuracy gains over end-to-end VLM approaches and 41.96–51.84% gains over specialized methods on the main task. It also claims lower inference cost and better computational efficiency than closed-source state-of-the-art VLMs. Code is available on GitHub.