Computer Science undergraduate · Robotics research

Yuxiao Zhu 朱煜肖

I am interested in building robots that understand the physical world, reason about tasks, and act reliably to help people. My work spans contact sensing for manipulation and reasoning and planning for multi-robot systems. Looking ahead, I hope to explore how learning through physical interaction can connect perception, reasoning, and action.

Embodied AI Robot Perception Reasoning and Planning Contact-Rich Manipulation

News

TacPilot is now available on arXiv.

CoPRE is now available on arXiv.

New RSS workshop abstract on belief-guided contact exploration under noisy proprioception.

Melding LLM and temporal logic is on arXiv and under review at npj Robotics.

Presenting DEXTER-LLM at IROS 2025 in Hangzhou.

Publications & Work in Progress

Selected papers

6 selected works

TacPilot paper teaser: four physical manipulation tasks on OpenArm, with event-triggered local tactile encoding and independently gated action corrections
Co-first Author arXiv preprint · 2026

TacPilot: Does Touch Need to Be Everywhere?

Yunzhen Zhang*, Yuxiao Zhu*, Jie Yang, Yufeng Miao, Bing Luo

A tactile residual framework that independently gates local tactile feature updates and bounded action corrections around a frozen visual policy. Evaluated in simulation and on OpenArm v2 across four contact-rich manipulation tasks.

CoPRE weak-contact detection on ARX L5 and G1, with protective retreat and contact-guided insertion demonstrations
Co-first Author ICRA 2027 · Under Review CoRL 2026 Workshop RSS 2026 Workshop · Early version

CoPRE: Improving Sensitivity in Proprioceptive Contact Detection for Low-Cost Robot Arms

Yuxiao Zhu*, Jinzhou Li*, Yifei Dong, Muhammad Suhail, Chunyuan Yang, Xinyuan Luo, Haoyu Li, Boyuan Chen, Xianyi Cheng

Detects weak contact through contact-free prediction and noise-weighted Jacobian aggregation. Validated on ARX L5 and G1, with retreat and insertion demos.

DEXTER-LLM
Co-first Author IROS 2025

DEXTER-LLM: Dynamic and Explainable Coordination of Multi-Robot Systems in Unknown Environments via Large Language Models

Yuxiao Zhu*, Junfeng Chen*, Xintong Zhang, Meng Guo, Zhongkui Li

Integrates LLM-based multi-stage reasoning, optimization-based task assignment, and adaptive human-in-the-loop verification for dynamic multi-robot task planning in unknown environments.

Research team and mobile robots in a large indoor human-swarm collaboration experiment
npj Robotics Under Review 2026

Melding LLM and temporal logic for reliable human-swarm collaboration in complex scenarios

Junfeng Chen, Yuxiao Zhu, An Zhuo, Xintong Zhang, Shuo Zhang, Guanghui Wen, Xiwang Dong, Meng Guo, Zhongkui Li

A neuro-symbolic framework that melds verifiable temporal-logic task planning with context-grounded LLM reasoning for reliable, low-overhead human-swarm collaboration in dynamic scenarios.

SLEI3D
T-ASE 2025

SLEI3D: Simultaneous Exploration and Inspection via Heterogeneous Fleets under Limited Communication

Junfeng Chen, Yuxiao Zhu, Xintong Zhang, Bing Luo, Meng Guo

A planning and coordination framework for heterogeneous multi-robot systems performing simultaneous 3D exploration, inspection, and real-time reporting under limited communication.

CoCoPlan
RA-L 2025

CoCoPlan: Adaptive Coordination and Communication for Multi-robot Systems in Dynamic and Unknown Environments

Xintong Zhang*, Junfeng Chen*, Yuxiao Zhu, Bing Luo, Meng Guo

Co-optimizes collaborative task planning and intermittent communication for multi-robot systems in dynamic environments under limited connectivity.

Experience

Research experience

Undergraduate Researcher

Duke Kunshan University / Edge Intelligence Lab

Supervised by Prof. Bing Luo and Yufeng Miao (helix robotics).

I am designing TacPilot's tactile residual framework and experiments to study when tactile features need updating and when touch should correct visual actions. I established the OpenArm manipulation platform through hardware integration and debugging.

Undergraduate Researcher

Duke University / DexLab

Supervised by Prof. Xianyi Cheng.

I led CoPRE from the initial concept through method design, implementation, and real-robot validation, developing weak-contact detection from built-in proprioception. I validated the method on ARX L5 and G1, with demonstrations of protective retreat and belief-guided insertion retries.

Visiting Undergraduate Researcher

Peking University

Hybrid online and in-person research supervised by Prof. Meng Guo and Prof. Zhongkui Li, with close mentorship from Dr. Junfeng Chen.

I co-designed DEXTER-LLM, CoCoPlan, and SLEI3D for multi-robot planning, coordination, and exploration. Across these projects, I led hardware integration and debugging for real-robot experiments and contributed to experimental design, evaluation, and manuscript writing.

Education

Academic background

Talks

Recent and upcoming presentations

DEXTER-LLM: Dynamic and Explainable Coordination of Multi-Robot Systems in Unknown Environments via Large Language Models

IROS 2025 Conference, Hangzhou International Expo Center, Hangzhou, China.

Presentation on dynamic task planning in unknown environments via LLM-based reasoning, optimization-based task assignment, and adaptive human-in-the-loop verification.

IROS 2025 presentation