Abstract: We propose RPR-LLaVA, a reinforcementaugmented program reasoning framework targeting multimodal mathematical tasks that integrate visual perception with symbolic reasoning. The model ...