Haoran Geng | 耿浩然

I am a Ph.D. student at Berkeley AI Research (BAIR), advised by Prof. Pieter Abbeel and Prof. Jitendra Malik. I also work closely with Prof. Alexei (Alyosha) Efros, Prof. Leonidas J. Guibas and Prof. Yue Wang. Previously, I was a visiting scholar at Stanford University through UGVR program. I received my Bachelor's degree with honors from Turing Class, Peking University. During my undergraduate years, I was honored to be advised by Prof. Leonidas J. Guibas and Dr. Siyuan Huang. I am also grateful to have grown up and studied with my twin brother Yiran Geng, which has been a truly unique and special experience for me.

Email  /  Google Scholar  /  Github /  Twitter /  WeChat

@misc{chen2025largevideoplanner,
title={Large Video Planner},
author={Boyuan Chen and Tianyuan Zhang and Haoran Geng and Kiwhan Song and William T. Freeman and Jitendra Malik and Russ Tedrake and Vincent Sitzmann and Yilun Du},
year={2025},
eprint={2512.15840},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={http://arxiv.org/abs/2512.15840},
}

LVP is a video foundation model for robotics that generates video plans, which we retarget into executable robot motions. We train an open video model at foundation-model scale for generative robotics planning.

@misc{zhang2025rodriguesnetworklearningrobot,
title={Rodrigues Network for Learning Robot Actions},
author={Jialiang Zhang and Haoran Geng and Yang You and Congyue Deng and Pieter Abbeel and Jitendra Malik and Leonidas Guibas},
year={2025},
eprint={2506.02618},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2506.02618}, }

We design the Rodrigues Network (RodriNet), a novel neural architecture specialized for processing actions.

@misc{heng2025vitacformerlearningcrossmodalrepresentation,
title={ViTacFormer: Learning Cross-Modal Representation for Visuo-Tactile Dexterous Manipulation},
author={Liang Heng and Haoran Geng and Kaifeng Zhang and Pieter Abbeel and Jitendra Malik},
year={2025},
eprint={2506.15953},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2506.15953},
}

We present ViTacFormer, a representation-learning approach that couples a cross-attention encoder to fuse high-resolution vision and touch with an autoregressive tactile prediction head that anticipates future contact signals.

@article{kuang2025skillblender,
title={SkillBlender: Towards Versatile Humanoid Whole-Body Loco-Manipulation via Skill Blending},
author={Kuang, Yuxuan and Geng, Haoran and Elhafsi, Amine and Do, Tan-Dzung and Abbeel, Pieter and Malik, Jitendra and Pavone, Marco and Wang, Yue},
journal={arXiv preprint arXiv:2506.09366},
year={2025},
}

We design the Rodrigues Network (RodriNet), a novel neural architecture specialized for processing actions.

@misc{seo2025fasttd3simplefastcapable,
title={FastTD3: Simple, Fast, and Capable Reinforcement Learning for Humanoid Control},
author={Younggyo Seo and Carmelo Sferrazza and Haoran Geng and Michal Nauman and Zhao-Heng Yin and Pieter Abbeel},
year={2025},
eprint={2505.22642},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2505.22642},
}

We introduce FastTD3, a simple, fast, and capable off-policy RL algorithm for humanoid control.

@misc{geng2025roboverseunifiedplatformdataset,
title={RoboVerse: Towards a Unified Platform, Dataset and Benchmark for Scalable and Generalizable Robot Learning},
author={Haoran Geng and Feishi Wang and Songlin Wei and Yuyang Li and Bangjun Wang and Boshi An and Charlie Tianyue Cheng and Haozhe Lou and Peihao Li and Yen-Jen Wang and Yutong Liang and Dylan Goetting and Chaoyi Xu and Haozhe Chen and Yuxi Qian and Yiran Geng and Jiageng Mao and Weikang Wan and Mingtong Zhang and Jiangran Lyu and Siheng Zhao and Jiazhao Zhang and Jialiang Zhang and Chengyang Zhao and Haoran Lu and Yufei Ding and Ran Gong and Yuran Wang and Yuxuan Kuang and Ruihai Wu and Baoxiong Jia and Carlo Sferrazza and Hao Dong and Siyuan Huang and Yue Wang and Jitendra Malik and Pieter Abbeel},
year={2025},
eprint={2504.18904},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2504.18904},
}

We propose RoboVerse, a comprehensive framework for advancing robotics through a simulation platform, synthetic dataset, and unified benchmarks.

@article{Physpart,
title = {PhysPart: Physically Plausible Part Completion for Interactable Objects},
author = {Luo*, Rundong and Geng*, Haoran and Deng, Congyue and Li, Puhao and Wang, Zan and Jia, Baoxiong and Guidbas, Leonidas and Huang, Siyuan},
journal = {International Conference on Robotics and Automation (ICRA)},
year = {2025},
url = {https://arxiv.org/abs/2408.13724},
}

PhysPart proposes a diffusion-based part generation model that utilizes geometric conditioning through classifier-free guidance and formulates physical constraints as a set of stability and mobility losses to guide the sampling process.

@misc{kuang2024ramretrievalbasedaffordancetransfer,
title={RAM: Retrieval-Based Affordance Transfer for Generalizable Zero-Shot Robotic Manipulation},
author={Yuxuan Kuang and Junjie Ye and Haoran Geng and Jiageng Mao and Congyue Deng and Leonidas Guibas and He Wang and Yue Wang}, year={2024},
eprint={2407.04689},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2407.04689},
}

RAM proposes a retrieve-and-transfer framework for zero-shot robotic manipulation, featuring generalizability across various objects, environments, and embodiments.

@inproceedings{
wei2024droma,
title={D3RoMa: Disparity Diffusion-based Depth Sensing for Material-Agnostic Robotic Manipulation},
author={Songlin Wei and Haoran Geng and Jiayi Chen and Congyue Deng and Cui Wenbo and Chengyang Zhao and Xiaomeng Fang and Leonidas Guibas and He Wang},
booktitle={8th Annual Conference on Robot Learning},
year={2024},
url={https://openreview.net/forum?id=7E3JAys1xO}
}

In this work, we propose D3RoMa, a learning-based depth estimation framework on stereo image pairs that predicts clean and accurate depth in diverse indoor scenes, even in the most challenging scenarios with translucent or specular surfaces

@misc{zhang2024dexgraspnet20learninggenerative,
title={DexGraspNet 2.0: Learning Generative Dexterous Grasping in Large-scale Synthetic Cluttered Scenes},
author={Jialiang Zhang and Haoran Liu and Danshi Li and Xinqiang Yu and Haoran Geng and Yufei Ding and Jiayi Chen and He Wang},
year={2024},
eprint={2410.23004},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2410.23004},
}

We synthesized a large-scale dexterous grasping dataset in cluttered scenes and designed a generative framework to learn grasping in the real world.

@INPROCEEDINGS{10802733,
author={Ding, Yufei and Geng, Haoran and Xu, Chaoyi and Fang, Xiaomeng and Zhang, Jiazhao and Wei, Songlin and Dai, Qiyu and Zhang, Zhizheng and Wang, He},
booktitle={2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
title={Open6DOR: Benchmarking Open-instruction 6-DoF Object Rearrangement and A VLM-based Approach},
year={2024},
volume={},
number={},
pages={7359-7366},
keywords={Three-dimensional displays;Benchmark testing;Propulsion;6-DOF;Real-time systems;Artificial intelligence;Intelligent robots;Synthetic data},
doi={10.1109/IROS58592.2024.10802733}
}

We present Open6DOR, a challenging and comprehensive benchmark for open-instruction 6-DoF object rearrangement tasks. Following this, we propose a zero-shot and robust method, Open6DORGPT, which proves effective in demanding simulation environments and real-world scenarios.

Simulately is a project where we gather useful information of robotics & physics simulators for cutting-edge robot learning research.

@misc{geng2023sage,
title={SAGE: Bridging Semantic and Actionable Parts for GEneralizable Articulated-Object Manipulation under Language Instructions},
author={Haoran Geng and Songlin Wei and Congyue Deng and Bokui Shen and He Wang and Leonidas Guibas},
year={2023},
eprint={2312.01307},
archivePrefix={arXiv},
primaryClass={cs.RO} }

We present SAGE, a framework bridging the understanding of semantic and actionable parts for generalizable manipulation of articulated objects using Large Language Models(LLMs) and Visual-Language Models(VLMs).

@misc{li2024ag2maniplearningnovelmanipulation,
title={Ag2Manip: Learning Novel Manipulation Skills with Agent-Agnostic Visual and Action Representations},
author={Puhao Li and Tengyu Liu and Yuyang Li and Muzhi Han and Haoran Geng and Shu Wang and Yixin Zhu and Song-Chun Zhu and Siyuan Huang},
year={2024},
eprint={2404.17521},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2404.17521},
}

We introduce Ag2Manip, which enables various robotic manipulation tasks without any domain-specific demonstrations. Ag2Manip also supports robust imitation learning of manipulation skills in the real world.

@article{shapellm24,
author = {Zekun Qi and Runpei Dong and Shaochen Zhang and Haoran Geng and Chunrui Han and Zheng Ge and He Wang and Li Yi and Kaisheng Ma},
title = {ShapeLLM: Universal 3D Object Understanding for Embodied Interaction},
journal = {CoRR},
volume = {abs/2402.17766},
year = {2024},
eprinttype = {arXiv},
eprint = {2402.17766},
}

We present ShapeLLM, the first 3D Multimodal Large Language Model (LLM) designed for embodied interaction, exploring a universal 3D object understanding with 3D point clouds and languages.

@misc{li2023manipllm,
title={ManipLLM: Embodied Multimodal Large Language Model for Object-Centric Robotic Manipulation},
author={Xiaoqi Li and Mingxu Zhang and Yiran Geng and Haoran Geng and Yuxing Long and Yan Shen and Renrui Zhang and Jiaming Liu and Hao Dong},
year={2023},
eprint={2312.16217},
archivePrefix={arXiv},
primaryClass={cs.CV}
}

We present ManipLLM, introducing an innovative approach for robot manipulation that leverages the robust reasoning capabilities of Multimodal Large Language Models (MLLMs) to enhance the stability and generalization of manipulation.

@misc{
you2023make,
title={Make a Donut: Language-Guided Hierarchical EMD-Space Planning for Zero-shot Deformable Object Manipulation},
author={Yang You and Bokui Shen and Congyue Deng and Haoran Geng and He Wang and Leonidas Guibas},
year={2023},
eprint={2311.02787},
archivePrefix={arXiv},
primaryClass={cs.RO}
}

In this work, we introduce a demonstration-free hierarchical planning approach capable of tackling intricate long-horizon tasks without necessitating any training

@article{wan2023unidexgrasp++,
title={UniDexGrasp++: Improving Dexterous Grasping Policy Learning via Geometry-aware Curriculum and Iterative Generalist-Specialist Learning},
author={Wan, Weikang and Geng, Haoran and Liu, Yun and Shan, Zikang and Yang, Yaodong and Yi, Li and Wang, He},
journal={arXiv preprint arXiv:2304.00464},
year={2023} }

We propose a novel, object-agnostic method for learning a universal policy for dexterous object grasping from realistic point cloud observations and proprioceptive information under a table-top setting.

@article{gong2023arnold,
title={ARNOLD: A Benchmark for Language-Grounded Task Learning With Continuous States in Realistic 3D Scenes},
author={Gong, Ran and Huang, Jiangyong and Zhao, Yizhou and Geng, Haoran and Gao, Xiaofeng and Wu, Qingyang and Ai, Wensi and Zhou, Ziheng and Terzopoulos, Demetri and Zhu, Song-Chun and Jia, Baoxiong and Huang, Siyuan},
journal={arXiv preprint arXiv:2304.04321},
year={2023}
}

We present ARNOLD, a benchmark that evaluates language-grounded task learning with continuous states in realistic 3D scenes.

@article{geng2022gapartnet,
title={GAPartNet: Cross-Category Domain-Generalizable Object Perception and Manipulation via Generalizable and Actionable Parts},
author={Geng, Haoran and Xu, Helin and Zhao, Chengyang and Xu, Chao and Yi, Li and Huang, Siyuan and Wang, He},
journal={arXiv preprint arXiv:2211.05272},
year={2022}
}

We propose to learn cross-category generalizable object perception and manipulation skills via Generalizable and Actionable Parts(GAPart), and present GAPartNet, a large-scale interactive dataset with rich part annotations.

@article{geng2023partmanip,
title={PartManip: Learning Cross-Category Generalizable Part Manipulation Policy from Point Cloud Observations},
author={Geng, Haoran and Li, Ziming and Geng, Yiran and Chen, Jiayi and Dong, Hao and Wang, He},
journal={arXiv preprint arXiv:2303.16958},
year={2023}
}

We introduce a large-scale, cross-category part-based object manipulation benchmark with tasks in realistic, vision-based settings and design a novel augmented state-to-vision distillation method for these challenging tasks.

@article{xu2023unidexgrasp,
title={UniDexGrasp: Universal Robotic Dexterous Grasping via Learning Diverse Proposal Generation and Goal-Conditioned Policy},
author={Xu, Yinzhen and Wan, Weikang and Zhang, Jialiang and Liu, Haoran and Shan, Zikang and Shen, Hao and Wang, Ruicheng and Geng, Haoran and Weng, Yijia and Chen, Jiayi and others},
journal={arXiv preprint arXiv:2303.00938},
year={2023}
}

We tackle the problem of learning universal robotic dexterous grasping from a point cloud observation under a table-top setting.

@article{geng2022end,
title={End-to-End Affordance Learning for Robotic Manipulation},
author={Geng, Yiran and An, Boshi and Geng, Haoran and Chen, Yuanpei and Yang, Yaodong and Dong, Hao},
journal={International Conference on Robotics and Automation (ICRA)},
year={2023}
}

In this study, we take advantage of visual affordance by using the contact information generated during the RL training process to predict contact maps of interest.

I was selected for the Ministry of Education Talent Program and conducted physics research at Nankai University during my high school years.