Research

Research Overview

Welcome to my research page! Here you can find my latest work and interests in multimodal large language models, reinforcement learning, and agents.

Reinforcement LearningAgentVision-Language ModelsPost-Training

Recent Publications

Reinforcement Learning

Uni-OPD: Unifying On-Policy Distillation with a Dual-Perspective RecipeWenjin Hou*, Shangpin Peng*, Weinong Wang, Zheng Ruan, et al.Arxiv 2026•paper•code

Uni-DPO: A Unified Paradigm for Dynamic Preference Optimization of LLMsShangpin Peng, Weinong Wang, Zhuotao Tian, Senqiao Yang, et al., Min ZhangICLR 2026•paper•code

Vision-Language Models

HunyuanOCR Technical ReportHunyuan Vision Team, Pengyuan Lyu, Xingyu Wan, Gengluo Li, Shangpin Peng, et al., Chengquan ZhangCore Contributor of Technical Report•paper•code•media

StrucTab: A Structured Optimization Framework for Table ParsingGengluo Li*, Shangpin Peng*, et al., Yu ZhouECCV 2026•paper•code

Mitigating Object Hallucinations via Sentence-Level Early InterventionShangpin Peng, Senqiao Yang, Li Jiang, Zhuotao TianICCV 2025•paper•code•media

Chronicles-OCR: A Cross-Temporal Perception Benchmark for the Evolutionary Trajectory of Chinese CharactersGengluo Li, Shangpin Peng, Xingyu Wan, Chengquan Zhang, et al., Han HuArxiv 2026•paper•code•media

ChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and FormatsShangpin Peng, Gengluo Li, Xingyu Wan, et al.Arxiv 2026•paper•code

Agent

PhoneBuddy: Training Open Models for Agentic Phone UseZhengyang Tang, et al., Shangpin Peng, et al., Chengquan Zhang, Han HuArxiv 2026•paper•project•code•media

PhoneHarness: Harnessing Phone-Use Agents through Mixed GUI, CLI, and Tool ActionsJason, et al., Shangpin Peng, et al., Chengquan Zhang, Han HuArxiv 2026•paper•project•code•media

PhoneWorld: Scaling Phone-Use Agent EnvironmentsZhengyang Tang, et al., Shangpin Peng, et al., Chengquan Zhang, Han HuArxiv 2026•paper•media

Safe, or Simply Incapable? Rethinking Safety Evaluation for Phone-Use AgentsZhengyang Tang, et al., Shangpin Peng, et al., Chengquan Zhang, Han HuArxiv 2026•paper•code

Useful Links

• Google Scholar
• GitHub Profile
• Personal Homepage

Feel free to explore the projects above or contact me for more information!