I am a Ph.D. candidate in Computer Science and Technology at the National University of Defense Technology (NUDT), affiliated with a research team led by an Academician of the Chinese Academy of Sciences. I am currently a visiting researcher at the Peking University–PsiBot Joint Laboratory, under the guidance of Prof. Yaodong Yang and Prof. Yuanpei Chen, and I am also a member of the PAIR Lab. My research focuses on Embodied AI and Large Models (LLM/VLM/VLA), spanning from high-level decision-making and reasoning (e.g., LLM-based agentic planning, Task and Motion Planning, Behavior Tree generation) to low-level motion control (e.g., Vision–Language–Action models, Real-World Reinforcement Learning).
News
- 2026.05: VLA-Arena accepted to ICML 2026
- 2025.04: SafeVLA accepted to NeurIPS 2025 as Spotlight (Top 3.2%)
- 2025.01: Academic homepage launched
Publications
MRBTP: Efficient Multi-Robot Behavior Tree Planning and Collaboration
Yishuai Cai, Xinglin Chen, Zhongxuan Cai, Yunxin Mao, Minglong Li, Wenjing Yang, Ji Wang
AAAI 2025 Oral · Top 4.6% | Citations: 12
- Proposes a multi-robot behavior tree planning algorithm with theoretical soundness and completeness guarantees. Achieves efficient multi-agent collaboration through cross-tree expansion and intention sharing.
@inproceedings{cai2025mrbtp,
title = {MRBTP: Efficient Multi-Robot Behavior Tree Planning and Collaboration},
author = {Cai, Yishuai and Chen, Xinglin and Cai, Zhongxuan and Mao, Yunxin and Li, Minglong and Yang, Wenjing and Wang, Ji},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)},
year = {2025},
note = {Oral, Top 4.6%}
}
Xinglin Chen*, Yishuai Cai*, Yunxin Mao, Minglong Li, Wenjing Yang, Weixia Xu, Ji Wang
- Proposes a two-stage behavior tree generation framework that integrates large language model understanding with optimal behavior tree expansion, generating semantically correct and task-feasible behavior trees from natural language instructions.
@inproceedings{chen2024llmobtea,
title = {Integrating Intent Understanding and Optimal Behavior Planning for Behavior Tree Generation from Human Instructions},
author = {Chen, Xinglin and Cai, Yishuai and Mao, Yunxin and Li, Minglong and Yang, Wenjing and Xu, Weixia and Wang, Ji},
booktitle = {Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI)},
year = {2024}
}
CABTO: Context-Aware Behavior Tree Grounding for Robot Manipulation
Yishuai Cai, Xinglin Chen, Yunxin Mao, Kun Hu, Minglong Li, Yaodong Yang, Yuanpei Chen
AAAI 2026 | Citations: -
- Proposes a context-aware behavior tree grounding approach that effectively grounds high-level semantic instructions into low-level robot-executable control trees, enabling real-world robot manipulation.
@inproceedings{cai2026cabto,
title = {CABTO: Context-Aware Behavior Tree Grounding for Robot Manipulation},
author = {Cai, Yishuai and Chen, Xinglin and Mao, Yunxin and Hu, Kun and Li, Minglong and Yang, Yaodong and Chen, Yuanpei},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)},
year = {2026}
}
DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation
Kefei Zhu, Fengshuo Bai, YuanHao Xiang, Yishuai Cai, Xinglin Chen, Ruochong Li, Xingtao Wang, Hao Dong, Yaodong Yang†, Xiaopeng Fan†, Yuanpei Chen†
NeurIPS 2025 Spotlight · Top 3.2% | Citations: 7
- Proposes a scalable self-improving data generation framework that efficiently generates high-quality dexterous manipulation training data through robot demonstration collection and self-supervised learning cycles.
@inproceedings{zhu2025dexflywheel,
title = {DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation},
author = {Zhu, Kefei and Bai, Fengshuo and Xiang, YuanHao and Cai, Yishuai and Chen, Xinglin and Li, Ruochong and Wang, Xingtao and Dong, Hao and Yang, Yaodong and Fan, Xiaopeng and Chen, Yuanpei},
booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
year = {2025},
note = {Spotlight, Top 3.2%}
}
SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning
Borong Zhang, Jiahao Li, Jiachen Shen, Yuhao Zhang, Yishuai Cai, Yuanpei Chen, Juntao Dai, Jiaming Ji, Yaodong Yang
NeurIPS 2025 Spotlight · Top 3.2% | Citations: 47
- Proposes an Integrated Safety Approach that systematically models safety requirements, actively elicits unsafe behaviors, and enforces safety constraints via constrained MDPs for safe reinforcement learning, achieving safety alignment of Vision-Language-Action models.
@inproceedings{zhang2025safevla,
title = {SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning},
author = {Zhang, Borong and Li, Jiahao and Shen, Jiachen and Zhang, Yuhao and Cai, Yishuai and Chen, Yuanpei and Dai, Juntao and Ji, Jiaming and Yang, Yaodong},
booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
year = {2025},
note = {Spotlight, Top 3.2%}
}
VLA-Arena: An Open-Source Framework for Benchmarking Vision-Language-Action Models
Borong Zhang, Jiahao Li, Jiachen Shen, Yishuai Cai, Yuhao Zhang, Yuanpei Chen, Juntao Dai, Jiaming Ji, Yaodong Yang
ICML 2026 | Citations: 15
- Proposes an open-source benchmark framework for systematic evaluation of Vision-Language-Action models across dimensions of safety, generalization, and task complexity, supporting fine-grained difficulty level design.
@inproceedings{zhang2026vlaarena,
title = {VLA-Arena: An Open-Source Framework for Benchmarking Vision-Language-Action Models},
author = {Zhang, Borong and Li, Jiahao and Shen, Jiachen and Cai, Yishuai and Zhang, Yuhao and Chen, Yuanpei and Dai, Juntao and Ji, Jiaming and Yang, Yaodong},
booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
year = {2026}
}
HBTP: Heuristic Behavior Tree Planning with Large Language Model Reasoning
Yishuai Cai, Xinglin Chen, Yunxin Mao, Minglong Li, Shaowu Yang, Wenjing Yang, Ji Wang
ICRA 2025 | Citations: 5
- Proposes a heuristic behavior tree planning framework that leverages large language model reasoning to generate effective heuristics, significantly improving search efficiency and solution quality.
@inproceedings{cai2025hbtp,
title = {HBTP: Heuristic Behavior Tree Planning with Large Language Model Reasoning},
author = {Cai, Yishuai and Chen, Xinglin and Mao, Yunxin and Li, Minglong and Yang, Shaowu and Yang, Wenjing and Wang, Ji},
booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
year = {2025}
}
RedVLA: Physical Red Teaming for Vision-Language-Action Models
Yuhao Zhang, Borong Zhang, Jingxuan Fan, Jiachen Shen, Yishuai Cai, Yaodong Yang, Jiaming Ji
arXiv 2026 | Citations: 2
- Proposes a physical red teaming framework for Vision-Language-Action models, systematically exposing safety risks and vulnerabilities of VLAs under adversarial physical perturbations.
Frequency-Aware Flow Matching for Continuous and Consistent Robotic Action Generation
Jianing Guo, Fangzheng Chen, Zihao Mao, Wong Lik Hang Kenny, Zhenhong Wu, Yu Li, Yishuai Cai, Yuanpei Chen, Yikun Ban, Kai Chen, Qi Dou, Yaodong Yang, Xianglong Liu, Huijie Zhao, Simin Li
arXiv 2026
- Proposes Frequency-Aware Flow Matching (FAFM), which maps action sequences to the frequency domain via DCT for flow matching and reconstructs continuous actions, while regularizing the first-order temporal derivative for temporally consistent and smooth motions; it adds no extra parameters, applies to flow-matching policies and VLA models, and improves success rates and smoothness in simulation and on a real-world Franka robot.
@article{guo2026fafm,
title = {Frequency-Aware Flow Matching for Continuous and Consistent Robotic Action Generation},
author = {Guo, Jianing and Chen, Fangzheng and Mao, Zihao and Wong, Lik Hang Kenny and Wu, Zhenhong and Li, Yu and Cai, Yishuai and Chen, Yuanpei and Ban, Yikun and Chen, Kai and Dou, Qi and Yang, Yaodong and Liu, Xianglong and Zhao, Huijie and Li, Simin},
journal = {arXiv preprint},
year = {2026}
}
Task2Morph: Differentiable Task-inspired Framework for Contact-Aware Robot Design
Yishuai Cai, Shaowu Yang, Minglong Li, Xinglin Chen, Yunxin Mao, Xiaodong Yi, Wenjing Yang
IROS 2023 | Citations: 7
@inproceedings{cai2023task2morph,
title = {Task2Morph: Differentiable Task-inspired Framework for Contact-Aware Robot Design},
author = {Cai, Yishuai and Yang, Shaowu and Li, Minglong and Chen, Xinglin and Mao, Yunxin and Yi, Xiaodong and Yang, Wenjing},
booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year = {2023}
}
Evolving Physical Instinct for Morphology and Control Co-Adaption
Xinglin Chen, Da Huang, Minglong Li, Yishuai Cai, Zhuoer Wen, Zhongxuan Cai, Wenjing Yang
IROS 2023 | Citations: 6
@inproceedings{chen2023instinct,
title = {Evolving Physical Instinct for Morphology and Control Co-Adaption},
author = {Chen, Xinglin and Huang, Da and Li, Minglong and Cai, Yishuai and Wen, Zhuoer and Cai, Zhongxuan and Yang, Wenjing},
booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year = {2023}
}
Zihui Song*, Yishuai Cai*, Yi Yu, Pengfei Xiang, Minglong Li, Zhiwei Liu, Mei Dai
Materials Genome Engineering Advances 2026 | IF: 13.9 | Citations: 1
- A survey on the new opportunities for integrating Large Language Models and Embodied Large Models with materials science, discussing AI-driven materials discovery, autonomous experimentation, and cross-modal scientific intelligence.
@article{song2026mgea,
title = {New Opportunities for the Integration of Artificial Intelligence with Materials Science: From Large Language Models to Embodied Large Models},
author = {Song, Zihui and Cai, Yishuai and Yu, Yi and Xiang, Pengfei and Li, Minglong and Liu, Zhiwei and Dai, Mei},
journal = {Materials Genome Engineering Advances},
year = {2026},
doi = {10.1002/mgea.70042}
}
Conflict-Aware Memory for Embodied Agents: Enhancing Vector Data Quality via Detection Rules
Kexin Ma, Haotian Wang, Shenglin Chen, Yishuai Cai, Huangyu, Ruochun Jin
ACL 2026
- Proposes Conflict Detection Rules (CDRs) to identify and manage data-quality issues in the vector memory of embodied agents, correcting the index structure and improving answer quality; CDR-equipped planners surpass the basic LLM planner by 15.25% and 14.25% on average in grammatical and interpretation accuracy, respectively.
@inproceedings{ma2026conflict,
title = {Conflict-Aware Memory for Embodied Agents: Enhancing Vector Data Quality via Detection Rules},
author = {Ma, Kexin and Wang, Haotian and Chen, Shenglin and Cai, Yishuai and Huangyu and Jin, Ruochun},
booktitle = {Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL)},
year = {2026}
}
Causal Effect Estimation under Network Interference with State Space Models
Yunxin Mao, Yishuai Cai, Zhaoye Li, Haotian Wang, Minglong Li, Wenjing Yang, Ji Wang
ICASSP 2026
- Introduces Multi-Hop Interference (MHI) to capture peer effects across multi-hop neighborhoods and proposes the Mamba-based Causal Interference Network (MCINet) with a Neighborhood Sampling Depth-First Ordering (NSDFO) algorithm, achieving more accurate causal effect estimation under network interference on both synthetic and real datasets.
@inproceedings{mao2026mcinet,
title = {Causal Effect Estimation under Network Interference with State Space Models},
author = {Mao, Yunxin and Cai, Yishuai and Li, Zhaoye and Wang, Haotian and Li, Minglong and Yang, Wenjing and Wang, Ji},
booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
year = {2026}
}
Honors and Awards
- Academic Scholarships
National Scholarship (Ph.D.) National Scholarship (Undergraduate) Outstanding Graduate of Zhejiang Province Zhejiang Provincial Government Scholarship Outstanding Student of NUDT
- 2023: First Prize, Robot Large Model and Embodied Intelligence Challenge (National)
- 2020: First Prize, Mathematical Contest in Modeling (MCM/ICM)
- 2019: First Prize, Asia-Pacific Mathematical Contest in Modeling
Education
- 2021–Present, Ph.D. | National University of Defense Technology | Computer Science and Technology
- 2024–Present, Joint Training | PKU–PsiBot Joint Laboratory | Advisors: Prof. Yaodong Yang, Prof. Yuanpei Chen