Deng Haoyuan

Deng Haoyuan (邓皓元)

I'm a first-year PhD student at Nanyang Technological University, supervised by Prof. Ziwei Wang in the PINE Lab. My research focuses on Embodied AI, with interests in Real-World Reinforcement Learning, Bimanual Manipulation, and Imitation Learning. I work on Robot Foundation Models to enable robots to generalize across diverse tasks.

Previously, I received my Bachelor's degree from Wuhan University, where I worked with Prof. Guangyi Yang. I have also worked as a Senior Scientist Intern at A*STAR in Singapore.

Email Scholar Linkedin Github CV

News

[2026-10] NORA-1.5 was accepted to the NeurIPS 2026 Robotics World Modeling Workshop.
[2026-09] One paper (UniIntervene) is accepted to CoRL 2026.
[2026-06] One paper (REMA) is accepted to RA-L.
[2026-04] One paper(E2HiL) is accepted to RA-L.
[2026-01] Two papers(VLA-Reasoner and MAP-VLA) are accepted to ICRA 2026.
[2025-11] Survey paper "A Survey on Reinforcement Learning of Vision-Language-Action Models for Robotic Manipulation" is available on TechRxiv.
[2025-10] We get the 2nd place in Challenge of Multimodal Robot Learning in InternUtopia and Real World and 1st place in simulation stage at IROS 2025.

Research

I'm interested in Real-World Reinforcement Learning, Bimanual Manipulation, Imitation Learning, and Robot Foundation Models.

Selected Publications * indicates equal contribution
	DynaHarness: A Dynamic Physical Harness for Self-Evolving Robot Agents Haoyuan Deng, Jiebin Liu, Tengxiao Zhang, Langning Yan, Hongye Cao, Ziwei Wang^† arXiv, 2026 project page / arXiv / Code A dynamic physical harness that pairs a frozen vision-language-action model with fast execution checks and uses failure evidence to improve its skills through regression-tested revisions.
	Facet-0: A Robotic Foundation Model for Contact-Rich Precise Manipulation Haoyuan Deng, Haichao Liu, Wenkai Guo, Yuan Ling, Zaijia Yang, Yuanjiang Xue, Haosheng Sun, Liangzi Wang, Ziwei Wang^† arXiv*, 2026 project page / arXiv / Dataset / Code A robotic foundation model combining vision, language, and force sensing with reinforcement learning for precise contact-rich assembly, trained on the 1,000-hour ManuFacet-1K dataset.
Recent Works
	UniIntervene: Agentic Intervention for Efficient Real-World Reinforcement Learning Haoyuan Deng, Yitong Gao, Yudong Lin, Haichao Liu, Zhenyu Wu, Ziwei Wang^† CoRL 2026 project page / arXiv / Code An autonomous intervention system that replaces frequent human corrections in robot RL, using a value-aware critic to detect unproductive rollouts and a memory-guided recovery policy to restore progress.
	E2HiL: Entropy-Guided Sample Selection for Efficient Real-World Human-in-the-Loop Reinforcement Learning Haoyuan Deng, Yudong Lin, Yuanjiang Xue, Haoyang Du, Qianzhun Wang, Boyang Zhou, Zhenyu Wu, Ziwei Wang^† RA-L 2026 project page / Paper / Code A sample-efficient human-in-the-loop RL framework using entropy-guided influence functions to actively select informative training samples.
	SafeBimanual: Diffusion-based Trajectory Optimization for Safe Bimanual Manipulation Haoyuan Deng, Wenkai Guo, Qianzhun Wang, Zhenyu Wu, Ziwei Wang^† CoRL 2025 project page / arXiv / Code A test-time trajectory optimization framework for any pre-trained diffusion-based bimanual manipulation policies, which imposes the safety constraints on bimanual actions.
	A Survey on Reinforcement Learning of Vision-Language-Action Models for Robotic Manipulation Haoyuan Deng, Zhenyu Wu, Haichao Liu, Wenkai Guo, Yuquan Xue, Ziyu Shan, Chuanrui Zhang, Bofang Jia, Yuan Ling, Guanxing Lu, Ziwei Wang^† TechRxiv*, 2025 Paper / GitHub The first comprehensive survey on reinforcement learning of vision-language-action models for robotic manipulation.
	VLA-Reasoner: Empowering Vision-Language-Action Models with Reasoning via Online Monte Carlo Tree Search Wenkai Guo, Guanxing Lu, Haoyuan Deng, Zhenyu Wu, Yansong Tang, Ziwei Wang^† ICRA 2026 project page / arXiv / Code A plug-in framework that augments vision-language-action models with long-horizon planning via online Monte Carlo Tree Search and kernel density estimation-based confidence sampling.
	AnyBimanual: Transferring Unimanual Policy for General Bimanual Manipulation Guanxing Lu^, Tengbo Yu^, Haoyuan Deng, Season Si Chen, Yansong Tang^†, Ziwei Wang ICCV 2025 project page / arXiv / Code A framework designed to transfer pre-trained unimanual manipulation policies to multi-task bimanual manipulation with few bimanual demonstrations.
	Action- and World Model-based Rewards Improve Vision-Language-Action Model Post-Training Chia-Yu Hung, Navonil Majumder, Haoyuan Deng, Liu Renhang, Yankang Ang, Amir Zadeh, Chuan Li, Dorien Herremans, Ziwei Wang, Soujanya Poria NeurIPS 2026 Robotics World Modeling Workshop arXiv NORA-1.5 is trained with preference rewards derived from world model predictions and action quality assessments, enabling more robust and generalizable robot manipulation.
Other Publications
	ManiGaussian++: Dynamic Gaussian Splatting for Multi-task Bimanual Manipulation Tengbo Yu^, Guanxing Lu^, Zaijia Yang^, Haoyuan Deng*, Season Si Chen, Jiwen Lu, Wenbo Ding, Guoqiang Hu, Yansong Tang^†, Ziwei Wang IROS 2025 arXiv / Code A novel framework that addresses the challenges of multi-task bimanual manipulation through hierarchical Gaussian world modeling.
	UniManip: General-Purpose Zero-Shot Robotic Manipulation with Agentic Operational Graph Haichao Liu, Yuanjiang Xue, Yuheng Zhou, Haoyuan Deng, Yinan Liang, Lihua Xie, Ziwei Wang^† Under Review arXiv A general-purpose zero-shot manipulation framework using an agentic operational graph to decompose and execute complex manipulation tasks without task-specific training data.
	Efficient Online Robotic Reinforcement Learning with Explicit Manifold Alignment Wenkai Guo, Peizhi Tang, Yukun Bai, Ziming Yan, Haoyuan Deng, Guanxing Lu, Shijie Li, Ziwei Wang RA-L 2026 Paper / Google Scholar REMA improves online robotic reinforcement learning by aligning exploration with a manifold of successful demonstrations, preserving informative near-success trials while penalizing uninformative deviations.
	MAP-VLA: Memory-Augmented Prompting for Vision-Language-Action Model in Robotic Manipulation Runhao Li, Wenkai Guo, Zhenyu Wu, Changyuan Wang, Haoyuan Deng, Zhenyu Weng, Yap-Peng Tan, Ziwei Wang^† ICRA 2026 arXiv A plug-and-play memory-augmented framework for frozen VLA models that constructs a memory library from demonstrations and retrieves relevant memories via trajectory similarity matching for long-horizon manipulation.

Experiences

A*STAR Advanced Remanufacturing and Technology Centre (ARTC)
Senior Scientist Intern
May 2024 – Sep 2024
Singapore

Academic Services

Journal Reviewer: IEEE RA-L
Conference Reviewer: ICRA, IROS, CoRL, NeurIPS 2026 RoboPAD

Honors and Awards

2023, Outstanding Undergraduate, Wuhan University
2023, Outstanding Undergraduate Overseas Exchange Scholarship, Wuhan University
2020, 2022, Merit Student (Top 5%), Wuhan University
2020, 2022, First Class Scholarship (Top 5%), Wuhan University
2022, Second prize of the 10th National University Student’s Opt-Sci-Tech Competition

Website Template.

News

Research

Selected Publications

Recent Works

Other Publications

Experiences

Academic Services

Honors and Awards