Pengyuan Guo

Pengyuan Guo

AgenticLab: A Real-world Robot Agent Platform that Can See, Think, and Act

Published:

Overview

In this work, we present AgenticLab, an open-source, real-world benchmark platform that integrates large vision-language models (VLMs) and large language models (LLMs) to enable zero-shot robotic manipulation through reasoning and embodied interaction.

Unlike existing simulation-based benchmarks, AgenticLab focuses on accessible hardware, Open Knowledge-based Models, and compositional agentic intelligence, allowing researchers to study how VLM/LLM models perform in real-world manipulation scenarios.

BibTeX

@misc{guo2026agenticlab,
  title={AgenticLab: A Real-World Robot Agent Platform that Can See, Think, and Act},
  author={Pengyuan Guo and Zhonghao Mai and Zhengtong Xu and Kaidi Zhang and Heng Zhang and Zichen Miao and Arash Ajoudani and Zachary Kingston and Qiang Qiu and Yu She},
  year={2026},
  eprint={2602.01662},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2602.01662},
}