Zhuoran (Jolia) Chen

Robotics PhD @ Georgia Tech  ·  NYU Shanghai '26  ·  Robotics

Looking for Summer 2027 internships in embodied AI, human-robot interaction, human-in-the-loop robot learning, and related areas. Feel free to reach out at zchen3135@gatech.edu!

I am a first-year Robotics PhD student at Georgia Institute of Technology, advised by Prof. Siddharth Karamcheti. I received my B.S. in Computer Science from NYU Shanghai in May 2026. My research broadly focuses on human-in-the-loop robot learning, exploring how robots can learn from, adapt to, and collaborate with people. I am particularly interested in making robot behavior more intuitive, adaptable, and aligned with human needs.

Previously, I worked with Prof. Shengjie Wang at NYU Shanghai on in-context learning for vision-language-action models in long-horizon manipulation tasks. At NYU's GRAIL Lab, I worked with Prof. Lerrel Pinto on enabling robots to learn force-sensitive manipulation from human touch. I also worked with Prof. Hongyi Wen at NYU Shanghai's MAPS Lab, focusing on personalized recommendation systems and human-in-the-loop image generation.

Outside the lab, I enjoy city walks, playing badminton, and capturing everyday moments through writing, photography, and painting.


News

Sep 2026 RUKA-v2 accepted to IEEE-RAS Humanoids 2026! 🎉
Sep 2026 Thrilled to be joining Georgia Institute of Technology as a Robotics PhD student this fall! 🤖
July 2026 TFP accepted to RSS 2026 SemRob Workshop! 🎉
May 2026 Graduated from NYU Shanghai with a B.S. in Computer Science. 🎓
Apr 2026 Awarded the Georgia Robotics Fellowship 2026. 🏆
Aug 2025 ImageGem accepted to ICCV 2025! 🎉
June 2025 Feel the Force accepted to RSS Workshop 2025! 🎉

Publications

* denotes equal contribution

Humanoid 2026

RUKA-v2: Tendon-Driven Open-Source Dexterous Hand with Wrist and Abduction for Robot Learning

Xinqi Lucas Liu*, Ruoxi Hu*, Alejandro Ojeda Olarte, Zhuoran Chen, Kenny Ma, Charles Cheng Ji, Lerrel Pinto, Raunaq Bhirangi, Irmak Guzey

IEEE-RAS Humanoids 2026 ICRA 2026 Dexterous Manipulation Workshop Workshop Spotlight
Paper Project Page

Lack of accessible and dexterous robot hardware has been a significant bottleneck to achieving human-level dexterity in robots. Last year, we released Ruka, a fully open-sourced, tendon-driven humanoid hand with 11 degrees of freedom — 2 per finger and 3 at the thumb — buildable for under $1,300. Despite these contributions, Ruka lacked two degrees of freedom essential for closely imitating human behavior: wrist mobility and finger adduction/abduction. In this paper, we introduce Ruka-v2: a fully open-sourced, tendon-driven humanoid hand featuring a decoupled 2-DOF parallel wrist and abduction/adduction at the fingers. The parallel wrist adds smooth, independent flexion/extension and radial/ulnar deviation, enabling manipulation in confined environments such as cabinets. Abduction enables motions such as grasping thin objects, in-hand rotation, and calligraphy. We evaluate Ruka-v2 against Ruka through user studies on teleoperated tasks, finding a 51.3% reduction in completion time and a 21.2% increase in success rate. We further demonstrate its full range of applications for robot learning: bimanual and single-arm teleoperation across 13 dexterous tasks, and autonomous policy learning on 3 tasks.

RSS 2026 TFP overview figure

TFP: Temporally Conditioned Memory-Fusion Policies for Visuomotor Learning

Yushen Liang*, Yue Peng*, Baosheng Jin*, Tianluo Zhang, Xinyu Zhang, Shuyi Zhou, Zhuoran Chen, Xinqi Liu, Shenji Wan

RSS 2026 SemRob Workshop
Paper Code

Vision–Language–Action (VLA) policies such as π0.5 and OpenVLA perform well on many manipulation tasks, but they are often reactive: the next action is predicted from the current observation, instruction, and proprioceptive state. This assumption breaks down in stage-dependent manipulation, where visually similar states may require different actions depending on latent task progress and previous interaction outcomes. We argue that such tasks require not only memory, but dynamics-aware belief updates: the policy should preserve task progress during stable or occluded phases and revise its belief near contact, release, or subgoal transitions. We introduce Temporally Conditioned Memory-Fusion Policies (TFP), a lightweight memory-action framework for VLA backbones. TFP maintains an episode-local task-progress belief with Liquid Time-Constant dynamics and injects the updated belief directly into the flow-matching action decoder through adaptive modulation. This lets temporally accumulated context shape the generated action chunk, rather than serving only as passive history context. With a 3.3B-parameter model, TFP improves the average success rate from 96.9% to 98.75% on LIBERO and from 91.4% to 93.77% on LIBERO-plus. On the memory-focused MIKASA ShellGameTouch diagnostic, TFP achieves success up to 75.0%. Mechanistic analyses show that write-gain changes near manipulation events are about 6 times larger than far non-event phases, and hidden-state interventions show that the belief causally modulates generated action chunks. These results suggest that compact, event-sensitive memory dynamics can improve VLA policies under occlusion, visual perturbation, and stage-dependent task structure.

2026
RSS 2025

Feel the Force: Contact-Driven Learning from Humans

Ademi Adeniji*, Zhuoran Chen*, Vincent Liu, Venkatesh Pattabiraman, Raunaq Bhirangi, Siddhant Haldar, Pieter Abbeel, Lerrel Pinto

RSS 2025 Learned Robot Representations (RoboReps) Workshop RSS 2025 Navigating Contact Dynamics in Robotics Workshop
Paper Project Page Code

Controlling fine-grained forces during manipulation remains a core challenge in robotics. While robot policies learned from robot-collected data or simulation show promise, they struggle to generalize across the diverse range of real-world interactions. Learning directly from humans offers a scalable solution, enabling demonstrators to perform skills in their natural embodiment and in everyday environments. However, visual demonstrations alone lack the information needed to infer precise contact forces. We present FeelTheForce (FTF): a robot learning system that models human tactile behavior to learn force-sensitive manipulation. Using a tactile glove to measure contact forces and a vision-based model to estimate hand pose, we train a closed-loop policy that continuously predicts the forces needed for manipulation. This policy is re-targeted to a Franka Panda robot with tactile gripper sensors using shared visual and action representations. At execution, a PD controller modulates gripper closure to track predicted forces — enabling precise, force-aware control. Our approach grounds robust low-level force control in scalable human supervision, achieving a 77% success rate across 5 force-sensitive manipulation tasks.

ICCV ImageGem cover

ImageGem: In-the-Wild Generative Image Interaction Dataset for Generative Model Personalization

Yuanhe Guo*, Linxi Xie*, Zhuoran Chen, Kangrui Yu, Ryan Po, Guandao Yang, Gordon Wetzstein, Hongyi Wen

ICCV 2025
Paper Project Page Code

We introduce ImageGem, a dataset for studying generative models that understand fine-grained individual preferences. We posit that a key challenge hindering the development of such a generative model is the lack of in-the-wild and fine-grained user preference annotations. Our dataset features real-world interaction data from 57K users, who collectively have built 242K customized LoRAs, written 3M text prompts, and created 5M generated images. With user preference annotations from our dataset, we were able to train better preference alignment models. In addition, leveraging individual user preference, we investigated the performance of retrieval models and a vision-language model on personalized image retrieval and generative model recommendation. Finally, we propose an end-to-end framework for editing customized diffusion models in a latent weight space to align with individual user preferences. Our results demonstrate that the ImageGem dataset enables, for the first time, a new paradigm for generative model personalization.

2025

Publications

* denotes equal contribution

ICRA 2026

RUKA-v2: Tendon-Driven Open-Source Dexterous Hand with Wrist and Abduction for Robot Learning

Xinqi Lucas Liu*, Ruoxi Hu*, Alejandro Ojeda Olarte, Zhuoran Chen, Kenny Ma, Charles Cheng Ji, Lerrel Pinto, Raunaq Bhirangi, Irmak Guzey

ICRA 2026 Dexterous Manipulation WorkshopSpotlight
Paper Project Page

Lack of accessible and dexterous robot hardware has been a significant bottleneck to achieving human-level dexterity in robots. Last year, we released Ruka, a fully open-sourced, tendon-driven humanoid hand with 11 degrees of freedom — 2 per finger and 3 at the thumb — buildable for under $1,300. Despite these contributions, Ruka lacked two degrees of freedom essential for closely imitating human behavior: wrist mobility and finger adduction/abduction. In this paper, we introduce Ruka-v2: a fully open-sourced, tendon-driven humanoid hand featuring a decoupled 2-DOF parallel wrist and abduction/adduction at the fingers. The parallel wrist adds smooth, independent flexion/extension and radial/ulnar deviation, enabling manipulation in confined environments such as cabinets. Abduction enables motions such as grasping thin objects, in-hand rotation, and calligraphy. We evaluate Ruka-v2 against Ruka through user studies on teleoperated tasks, finding a 51.3% reduction in completion time and a 21.2% increase in success rate. We further demonstrate its full range of applications for robot learning: bimanual and single-arm teleoperation across 13 dexterous tasks, and autonomous policy learning on 3 tasks.

RSS 2026 TFP overview figure

TFP: Temporally Conditioned Memory-Fusion Policies for Visuomotor Learning

Yushen Liang*, Yue Peng*, Baosheng Jin*, Tianluo Zhang, Xinyu Zhang, Shuyi Zhou, Zhuoran Chen, Xinqi Liu, Shenji Wan

RSS 2026 SemRob Workshop
Paper Code

Vision–Language–Action (VLA) policies such as π0.5 and OpenVLA perform well on many manipulation tasks, but they are often reactive: the next action is predicted from the current observation, instruction, and proprioceptive state. This assumption breaks down in stage-dependent manipulation, where visually similar states may require different actions depending on latent task progress and previous interaction outcomes. We argue that such tasks require not only memory, but dynamics-aware belief updates: the policy should preserve task progress during stable or occluded phases and revise its belief near contact, release, or subgoal transitions. We introduce Temporally Conditioned Memory-Fusion Policies (TFP), a lightweight memory-action framework for VLA backbones. TFP maintains an episode-local task-progress belief with Liquid Time-Constant dynamics and injects the updated belief directly into the flow-matching action decoder through adaptive modulation. This lets temporally accumulated context shape the generated action chunk, rather than serving only as passive history context. With a 3.3B-parameter model, TFP improves the average success rate from 96.9% to 98.75% on LIBERO and from 91.4% to 93.77% on LIBERO-plus. On the memory-focused MIKASA ShellGameTouch diagnostic, TFP achieves success up to 75.0%. Mechanistic analyses show that write-gain changes near manipulation events are about 6 times larger than far non-event phases, and hidden-state interventions show that the belief causally modulates generated action chunks. These results suggest that compact, event-sensitive memory dynamics can improve VLA policies under occlusion, visual perturbation, and stage-dependent task structure.

2026
RSS 2025

Feel the Force: Contact-Driven Learning from Humans

Ademi Adeniji*, Zhuoran Chen*, Vincent Liu, Venkatesh Pattabiraman, Raunaq Bhirangi, Siddhant Haldar, Pieter Abbeel, Lerrel Pinto

RSS 2025 Learned Robot Representations (RoboReps) Workshop RSS 2025 Navigating Contact Dynamics in Robotics Workshop
Paper Project Page Code

Controlling fine-grained forces during manipulation remains a core challenge in robotics. While robot policies learned from robot-collected data or simulation show promise, they struggle to generalize across the diverse range of real-world interactions. Learning directly from humans offers a scalable solution, enabling demonstrators to perform skills in their natural embodiment and in everyday environments. However, visual demonstrations alone lack the information needed to infer precise contact forces. We present FeelTheForce (FTF): a robot learning system that models human tactile behavior to learn force-sensitive manipulation. Using a tactile glove to measure contact forces and a vision-based model to estimate hand pose, we train a closed-loop policy that continuously predicts the forces needed for manipulation. This policy is re-targeted to a Franka Panda robot with tactile gripper sensors using shared visual and action representations. At execution, a PD controller modulates gripper closure to track predicted forces — enabling precise, force-aware control. Our approach grounds robust low-level force control in scalable human supervision, achieving a 77% success rate across 5 force-sensitive manipulation tasks.

ICCV 2025 ImageGem cover

ImageGem: In-the-Wild Generative Image Interaction Dataset for Generative Model Personalization

Yuanhe Guo*, Linxi Xie*, Zhuoran Chen, Kangrui Yu, Ryan Po, Guandao Yang, Gordon Wetzstein, Hongyi Wen

ICCV 2025
Paper Project Page Code

We introduce ImageGem, a dataset for studying generative models that understand fine-grained individual preferences. We posit that a key challenge hindering the development of such a generative model is the lack of in-the-wild and fine-grained user preference annotations. Our dataset features real-world interaction data from 57K users, who collectively have built 242K customized LoRAs, written 3M text prompts, and created 5M generated images. With user preference annotations from our dataset, we were able to train better preference alignment models. In addition, leveraging individual user preference, we investigated the performance of retrieval models and a vision-language model on personalized image retrieval and generative model recommendation. Finally, we propose an end-to-end framework for editing customized diffusion models in a latent weight space to align with individual user preferences. Our results demonstrate that the ImageGem dataset enables, for the first time, a new paradigm for generative model personalization.

2025

Experience

Research Experience

Sep 2025 – Present
Research Assistant  ·  NYU Shanghai
  • VLA models + in-context learning for long-horizon robotic tasks
Dec 2024 – Aug 2025
Research Assistant  ·  NYU GRAIL Lab (New York)
  • Feel the Force (co-first author)
  • RUKA-v2 teleoperation system
Jan 2024 – May 2025
Research Assistant  ·  NYU Shanghai MAPS Lab
  • ImageGem (second author)
  • Human-in-the-Loop Image Generation user study

Teaching Experience

Mar 2026 – Present
Teaching Assistant  ·  NYU Shanghai
SHBI-GB 7318 Artificial Intelligence for Business
Jan 2026 – Present
Learning Assistant  ·  NYU Shanghai
CSCI-SHU 210 Data Structures
CSCI-SHU 11 Introduction to Computer Programming
Sep 2022 – May 2023
Teaching Tutor  ·  Stepping Stone Program

Awards & Honors

Georgia Robotics Fellowship 2026
China National Scholarship 中国国家奖学金 2025
NYU Shanghai Recognition Award 2025
NYU Shanghai Outstanding Capstone Poster Award 2025
NYU Shanghai Dean Undergraduate Research Funding 2023, 2024
NYU Shanghai Dean's Honors List 2023 – 2025

Course Projects

Style Unlearning: Parameter-Efficient Modules for Efficient Style Removal

CSCI-GA 2271 Computer Vision  ·  Prof. Saining Xie

NYU  ·  Fall 2024

Compared W2W framework vs LoRA Negation for style unlearning; verified with CLIP + VLM metrics.

Computer Graphics Final Project

CSCI-GA 2270 Computer Graphics  ·  Prof. Ken Perlin

NYU  ·  Fall 2024

Interactive 3D web app using raw WebGL — 3D modeling, animation, real-time rendering.

Evaluation Methods on Text Summarization

CSCI-SHU 376 Natural Language Processing  ·  Prof. Chen Zhao

NYU Shanghai  ·  Spring 2024

Evaluated GPT-3.5/4 summarization with ROUGE, BERTScore, and human assessment.

Audio Classification using Deep Learning

CSCI-SHU 360 Machine Learning  ·  Prof. Yik-Cheung Tam

NYU Shanghai  ·  Spring 2024

End-to-end audio classification with Mel spectrograms, MFCCs, CNN models — 79.6% test accuracy.

Multi-function Chat System with GUI

CSCI-SHU 101 Introduction to Computer and Data Science  ·  Prof. Xianbin Gu

NYU Shanghai  ·  Fall 2023

GUI chat app with RSA encryption, login, ELO image rating, PyGame Flappy Bird module.

Misc

The non-academic side of me.

✝️ | ENFJ

Interests

🚶 City walk 🏸 Badminton 🏃 Running 🎨 Painting

Speaking Languages

English — Fluent Cantonese — Native Mandarin — Native

Portrait

Portrait painted by my sister Another portrait painted by my sister

painted by my sister ♥

揾我倾偈 (wan2 ngo5 keng1 gai2)  ·  Open to Chat

I set aside 1 hour each week for anyone who wants to talk — research directions, PhD applications, life abroad, robotics, or genuinely anything. No agenda needed. Send me an email and we'll find a time 😄.

🏸 🚶‍♀️‍➡️  ·  Open to badminton / citywalk

If you are in Atlanta and want to play badminton, or just walk and talk — send me an email. Always down :)