Publications

* Equal contribution   † Corresponding author

2026

Fewer Tokens, Better Action: GPT-6 Astra Robot Agents with 14% Higher Success but 65% Fewer Tokens

Ruiyang Si, Jianxin Bi, Shunyu Yang, Rui Ni, Wenbo Huang, Qiang Wang, Shulong Jiang, Yintianrun Zhang, Feixiang Ruan, Kewei Zhang, Duomin Wang, Xiuyu Li, Haiwen Feng, Zhen Dong, Daquan Zhou†

Technical Report | arXiv | Project | Code | Model

An efficient code-execution framework for robot-use agents: the agent chains robot primitives in Python cells, raising success from 63.1% to 71.7% with 65% fewer input tokens. Achieves state-of-the-art on RoboDojo (51.6% success, 19.4 points above the previous best on the public leaderboard), and our GRPO-trained 4B Coder generalizes to unseen tasks (18.8% → 44.1%), outperforming Qwen3.5-27B.

2025

Object-AVEdit: An Object-level Audio-Visual Editing Model

Youquan Fu*, Ruiyang Si*, Hongfa Wang, Dongzhan Zhou, Jiacheng Sun, Ping Luo, Di Hu†, Hongyuan Zhang†, Xuelong Li†

Preprint | arXiv

We train an audio generation model whose text tokens are explicitly aligned with sounding objects, and build on it the first model for object-level audio-visual editing, supporting object addition, replacement and removal across both audio and video.