Active Zero: Self-Evolving Vision-Language Models through Active Environment Exploration

2026-02-11

AI Summary

Existing self-play approaches for vision-language models are constrained by static image datasets, limiting their ability to actively acquire visual information aligned with their evolving capabilities and thus hindering learning efficiency. This work proposes a closed-loop system comprising three co-evolving agents—Searcher, Questioner, and Solver—that introduces active environmental exploration into the self-evolution process of vision-language models for the first time. By integrating open-world image retrieval with adaptive task generation, the framework establishes a frontier-driven automated curriculum learning mechanism. Evaluated on Qwen2.5-VL-7B-Instruct, the method achieves an average accuracy of 53.97% (+5.7%) on reasoning tasks and 59.77% (+3.9%) on general understanding across 12 benchmarks, significantly outperforming current self-play methods.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Self-play has enabled large language models to autonomously improve through self-generated challenges. However, existing self-play methods for vision-language models rely on passive interaction with static image collections, resulting in strong dependence on initial datasets and inefficient learning. Without the ability to actively seek visual data tailored to their evolving capabilities, agents waste computational effort on samples that are either trivial or beyond their current skill level. To address these limitations, we propose Active-Zero, a framework that shifts from passive interaction to active exploration of visual environments. Active-Zero employs three co-evolving agents: a Searcher that retrieves images from open-world repositories based on the model's capability frontier, a Questioner that synthesizes calibrated reasoning tasks, and a Solver refined through accuracy rewards. This closed loop enables self-scaffolding auto-curricula where the model autonomously constructs its learning trajectory. On Qwen2.5-VL-7B-Instruct across 12 benchmarks, Active-Zero achieves 53.97 average accuracy on reasoning tasks (5.7% improvement) and 59.77 on general understanding (3.9% improvement), consistently outperforming existing self-play baselines. These results highlight active exploration as a key ingredient for scalable and adaptive self-evolving vision-language systems.

Problem

• vision-language models
• self-play
• active exploration
• auto-curriculum
• environment interaction

Innovation

• active exploration
• self-evolving
• vision-language models
• auto-curriculum
• co-evolving agents