UIShift: Enhancing VLM-based GUI Agents through Self-supervised Reinforcement Learning

• URL: http://arxiv.org/abs/2505.12493v1
• Date: Sun, 18 May 2025 16:34:30 GMT
• Title: UIShift: Enhancing VLM-based GUI Agents through Self-supervised Reinforcement Learning
• Authors: Longxi Gao, Li Zhang, Mengwei Xu,
• Abstract summary: Training effective Vision Language Models (VLMs) for GUI agents typically relies on supervised fine-tuning (SFT) over large-scale annotated datasets.<n>We propose a self-supervised inverse dynamics task to enable VLMs to learn from GUI transition pairs by inferring the action that caused that transition.<n>We propose UI-shift, a framework for enhancing VLM-based GUI agents through self-supervised reinforcement learning.
• Score: 4.18969040567543
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Training effective Vision Language Models (VLMs) for GUI agents typically relies on supervised fine-tuning (SFT) over large-scale annotated datasets, where the collection process is labor-intensive and error-prone. In this work, we propose a self-supervised inverse dynamics task to enable VLMs to learn from GUI transition pairs by inferring the action that caused that transition. This training task offers two advantages: (1) It enables VLMs to ignore variations unrelated to user actions (e.g., background refreshes, ads) and to focus on true affordances such as buttons and input fields within complex GUIs. (2) The training data can be easily obtained from existing GUI trajectories without requiring human annotation, and it can be easily scaled through automatic offline exploration. Using this training task, we propose UI-shift, a framework for enhancing VLM-based GUI agents through self-supervised reinforcement learning (RL). With only 2K training samples sourced from existing datasets, two VLMs -- Qwen2.5-VL-3B and Qwen2.5-VL-7B -- trained with UI-Shift achieve competitive or superior performance on grounding tasks (ScreenSpot-series benchmarks) and GUI automation tasks (AndroidControl), compared to SFT baselines and GUI-specific models that explicitly elicit reasoning abilities during RL. Our findings suggest a potential direction for enhancing VLMs for GUI agents by leveraging more self-supervised training data in the future.

Related papers

• MobileGUI-RL: Advancing Mobile GUI Agent through Reinforcement Learning in Online Environment [63.62778707277929]
  MobileGUI-RL is a scalable framework that trains GUI agent in online environment.<n>It synthesizes a curriculum of learnable tasks through self-exploration and filtering.<n>It adapts GRPO to GUI navigation with trajectory-aware advantages and composite rewards.
  arXiv  Detail & Related papers  (2025-07-08T07:07:53Z)
• ZeroGUI: Automating Online GUI Learning at Zero Human Cost [75.21128388931945]
  We propose ZeroGUI, a scalable, online learning framework for automating GUI Agent training at Zero human cost.<n>Specifically, ZeroGUI integrates (i) VLM-based automatic task generation to produce diverse training goals from the current environment state, (ii) VLM-based automatic reward estimation to assess task success without hand-crafted evaluation functions, and (iii) two-stage online reinforcement learning to continuously interact with and learn from GUI environments.
  arXiv  Detail & Related papers  (2025-05-29T17:59:51Z)
• Guiding VLM Agents with Process Rewards at Inference Time for GUI Navigation [101.09478572153239]
  We propose an approach that guides VLM agents with process supervision by a reward model during GUI navigation and control at inference time.<n>This guidance allows the VLM agent to optimize actions at each inference step, thereby improving performance in both static and dynamic environments.
  arXiv  Detail & Related papers  (2025-04-22T17:52:42Z)
• TRISHUL: Towards Region Identification and Screen Hierarchy Understanding for Large VLM based GUI Agents [0.6827423171182154]
  TRISHUL is a training-free framework that enhances generalist LVLMs for holistic GUI comprehension.<n>Our results demonstrate TRISHUL's superior performance in action grounding across the ScreenSpot, VisualWebBench, AITW, and Mind2Web datasets.<n>For GUI referring, TRISHUL surpasses the ToL agent on the ScreenPR benchmark, setting a new standard for robust and adaptable GUI comprehension.
  arXiv  Detail & Related papers  (2025-02-12T09:12:30Z)
• UI-TARS: Pioneering Automated GUI Interaction with Native Agents [58.18100825673032]
  This paper introduces UI-TARS, a native GUI agent model that solely perceives the screenshots as input and performs human-like interactions.<n>In the OSWorld benchmark, UI-TARS achieves scores of 24.6 with 50 steps and 22.7 with 15 steps, outperforming Claude (22.0 and 14.9 respectively)
  arXiv  Detail & Related papers  (2025-01-21T17:48:10Z)
• InfiGUIAgent: A Multimodal Generalist GUI Agent with Native Reasoning and Reflection [38.833925781308665]
  We introduce textitInfiGUIAgent, an MLLM-based GUI Agent trained with a two-stage supervised fine-tuning pipeline.<n> Stage 1 enhances fundamental skills such as GUI understanding and grounding, while Stage 2 integrates hierarchical reasoning and expectation-reflection reasoning skills.<n>textitInfiGUIAgent achieves competitive performance on several GUI benchmarks.
  arXiv  Detail & Related papers  (2025-01-08T15:45:21Z)
• Aguvis: Unified Pure Vision Agents for Autonomous GUI Interaction [69.57190742976091]
  Aguvis is a vision-based framework for autonomous GUI agents.<n>It standardizes cross-platform interactions and incorporates structured reasoning via inner monologue.<n>It achieves state-of-the-art performance across offline and real-world online benchmarks.
  arXiv  Detail & Related papers  (2024-12-05T18:58:26Z)
• ShowUI: One Vision-Language-Action Model for GUI Visual Agent [80.50062396585004]
  Building Graphical User Interface (GUI) assistants holds significant promise for enhancing human workflow productivity. We develop a vision-language-action model in digital world, namely ShowUI, which features the following innovations. ShowUI, a lightweight 2B model using 256K data, achieves a strong 75.1% accuracy in zero-shot screenshot grounding.
  arXiv  Detail & Related papers  (2024-11-26T14:29:47Z)
• GUICourse: From General Vision Language Models to Versatile GUI Agents [75.5150601913659]
  We contribute GUICourse, a suite of datasets to train visual-based GUI agents. First, we introduce the GUIEnv dataset to strengthen the OCR and grounding capabilities of VLMs. Then, we introduce the GUIAct and GUIChat datasets to enrich their knowledge of GUI components and interactions.
  arXiv  Detail & Related papers  (2024-06-17T08:30:55Z)
• GUI-World: A Video Benchmark and Dataset for Multimodal GUI-oriented Understanding [73.9254861755974]
  This paper introduces a new dataset, termed GUI-World, which features meticulously crafted Human-MLLM annotations.<n>We evaluate the capabilities of current state-of-the-art MLLMs, including Image LLMs and Video LLMs, in understanding various types of GUI content.
  arXiv  Detail & Related papers  (2024-06-16T06:56:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.