CODA: Coordinating the Cerebrum and Cerebellum for a Dual-Brain Computer Use Agent with Decoupled Reinforcement Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sun, Zeyi, Cao, Yuhang, Liang, Jianze, Sun, Qiushi, Liu, Ziyu, Zhang, Zhixiong, Zang, Yuhang, Dong, Xiaoyi, Chen, Kai, Lin, Dahua, Wang, Jiaqi
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912556635914240
author Sun, Zeyi
Cao, Yuhang
Liang, Jianze
Sun, Qiushi
Liu, Ziyu
Zhang, Zhixiong
Zang, Yuhang
Dong, Xiaoyi
Chen, Kai
Lin, Dahua
Wang, Jiaqi
author_facet Sun, Zeyi
Cao, Yuhang
Liang, Jianze
Sun, Qiushi
Liu, Ziyu
Zhang, Zhixiong
Zang, Yuhang
Dong, Xiaoyi
Chen, Kai
Lin, Dahua
Wang, Jiaqi
contents Autonomous agents for Graphical User Interfaces (GUIs) face significant challenges in specialized domains such as scientific computing, where both long-horizon planning and precise execution are required. Existing approaches suffer from a trade-off: generalist agents excel at planning but perform poorly in execution, while specialized agents demonstrate the opposite weakness. Recent compositional frameworks attempt to bridge this gap by combining a planner and an actor, but they are typically static and non-trainable, which prevents adaptation from experience. This is a critical limitation given the scarcity of high-quality data in scientific domains. To address these limitations, we introduce CODA, a novel and trainable compositional framework that integrates a generalist planner (Cerebrum) with a specialist executor (Cerebellum), trained via a dedicated two-stage pipeline. In the first stage, Specialization, we apply a decoupled GRPO approach to train an expert planner for each scientific application individually, bootstrapping from a small set of task trajectories. In the second stage, Generalization, we aggregate all successful trajectories from the specialized experts to build a consolidated dataset, which is then used for supervised fine-tuning of the final planner. This equips CODA with both robust execution and cross-domain generalization. Evaluated on four challenging applications from the ScienceBoard benchmark, CODA significantly outperforms baselines and establishes a new state of the art among open-source models.
format Preprint
id arxiv_https___arxiv_org_abs_2508_20096
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CODA: Coordinating the Cerebrum and Cerebellum for a Dual-Brain Computer Use Agent with Decoupled Reinforcement Learning
Sun, Zeyi
Cao, Yuhang
Liang, Jianze
Sun, Qiushi
Liu, Ziyu
Zhang, Zhixiong
Zang, Yuhang
Dong, Xiaoyi
Chen, Kai
Lin, Dahua
Wang, Jiaqi
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Autonomous agents for Graphical User Interfaces (GUIs) face significant challenges in specialized domains such as scientific computing, where both long-horizon planning and precise execution are required. Existing approaches suffer from a trade-off: generalist agents excel at planning but perform poorly in execution, while specialized agents demonstrate the opposite weakness. Recent compositional frameworks attempt to bridge this gap by combining a planner and an actor, but they are typically static and non-trainable, which prevents adaptation from experience. This is a critical limitation given the scarcity of high-quality data in scientific domains. To address these limitations, we introduce CODA, a novel and trainable compositional framework that integrates a generalist planner (Cerebrum) with a specialist executor (Cerebellum), trained via a dedicated two-stage pipeline. In the first stage, Specialization, we apply a decoupled GRPO approach to train an expert planner for each scientific application individually, bootstrapping from a small set of task trajectories. In the second stage, Generalization, we aggregate all successful trajectories from the specialized experts to build a consolidated dataset, which is then used for supervised fine-tuning of the final planner. This equips CODA with both robust execution and cross-domain generalization. Evaluated on four challenging applications from the ScienceBoard benchmark, CODA significantly outperforms baselines and establishes a new state of the art among open-source models.
title CODA: Coordinating the Cerebrum and Cerebellum for a Dual-Brain Computer Use Agent with Decoupled Reinforcement Learning
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2508.20096