👋 About Me
I am currently an M.S. student in Mathematics at Liaoning Technical University.
My research interests include Machine Learning, Reinforcement Learning, and Large Language Models.
🔥 News
2023—present，I am collaborating with Haonan Wang on various projects for AI domain.
🎓 Education
Liaoning Technical University， M.S. in Mathematics, 2025–Present
Liaoning Technical University， B.S. in Information and Computing Science, 2019–2023
🏭 Working experience
SpringTengAI, Co-creator and collaborator， 2025~Present
Beijing CONCO Technology Co., Ltd., Computer Vision Algorithm Engineer, 2023.7~2024.1
I worked on AI-driven computer vision tasks, including object detection, image recognition, image segmentation, and embedded system deployment, covering the end-to-end development process from model design to real-world deployment.
📝 Papers
[1] Bytesized32refactored: Towards an extensible interactive text games corpus for llm world modeling and evaluation
Haonan Wang, Junfeng Sun, Xingdi Yuan, Ruoyao Wang, Ziang Xiao
When Language Meets Games Workshop, EMNLP 2025.
[Paper] [Code]
[2] Research on geometric figure classification algorithm based on Deep Learning
Ruiyang Wang, Haonan Wang, Junfeng Sun, Mingjia Zhao, Meng Liu
Scientific Journal of Intelligent Systems Research(IEEE), 2022.
[Paper]
[Ongoing Project] Any-Step Dynamics Models for Long-Horizon offline Goal-Conditioned Reinforcement Learning
In Offline Goal-Conditioned Reinforcement Learning (Offline GCRL), long-horizon reasoning tasks remain challenging. This project proposed reachability-aware long-horizon prediction. Although Any-Step Dynamics Models can directly predict future goals at arbitrary horizons while avoiding recursive prediction error, the generated goals may not be reachable from the current state within the specified horizon, especially in environments with complex topology.
To address this issue, we verify the horizon-consistent reachability of generated goals using both explicit state-space constraints and implicit temporal structure. In particular, we leverage TempDATA-style temporal representations to capture latent reachability information that cannot be reliably inferred from geometric distance alone.
We are currently evaluating our method on long-horizon navigation benchmarks, including PointMaze, AntMaze, and HumanoidMaze, with a particular focus on whether reachability-aware prediction can improve the reliability of long-horizon goal generation and overall goal-reaching performance.
🏅 Honors and Awards
2025
- Second Prize, “Huawei Cup” China Graduate Mathematical Contest in Modeling.
- Third Prize, MathorCup Graduate Mathematical Modeling Challenge。
2022
- National Inspirational Scholarship, Ministry of Education of China.
- Third Prize, Liaoning Provincial “Shuo Ri Cup” Undergraduate Computer Design Competition.
2021
- Third Prize, Chinese Collegiate Computing Competition.
- Third Prize, MathorCup University Mathematical Modeling Challenge.
- First Prize, Liaoning Provincial “Shuo Ri Cup” Undergraduate Computer Design Competition.
- First Prize, Liaoning Provincial Agricultural Economic Modeling Competition.