ProgRM: Build Better GUI Agents with Progress Rewards

TL;DR

ProgRM introduces a novel progress reward model that provides dense, intermediate feedback for training GUI agents, improving their performance over existing reward models and proprietary LLMs.

Contribution

The paper proposes ProgRM, a new progress reward model with an LCS-based self-annotation algorithm for fine-grained feedback in GUI agent training.

Findings

ProgRM-trained actors outperform ORM-based and proprietary LLMs.

The LCS-based annotation effectively discovers key trajectory steps.

Extensive experiments validate the effectiveness of ProgRM.

Abstract

LLM-based (Large Language Model) GUI (Graphical User Interface) agents can potentially reshape our daily lives significantly. However, current LLM-based GUI agents suffer from the scarcity of high-quality training data owing to the difficulties of trajectory collection and reward annotation. Existing works have been exploring LLMs to collect trajectories for imitation learning or to offer reward signals for online RL training. However, the Outcome Reward Model (ORM) used in existing works cannot provide finegrained feedback and can over-penalize the valuable steps in finally failed trajectories. To this end, we propose Progress Reward Model (ProgRM) to provide dense informative intermediate rewards by predicting a task completion progress for each step in online training. To handle the challenge of progress reward label annotation, we further design an efficient LCS-based (Longest Common Subsequence) self-annotation algorithm to discover the key steps in trajectories and assign progress labels accordingly. ProgRM is evaluated with extensive experiments and analyses. Actors trained with ProgRM outperform leading proprietary LLMs and ORM-trained actors, illustrating the effectiveness of ProgRM. The codes for experiments will be made publicly available upon acceptance.