EAGLE-360: Embodied Active Global-to-Local Exploration in 360$^\circ$

2026-07-02

AI Summary

Existing multimodal large language models struggle to model polar coordinate distortion and cylindrical continuous topology in 360° panoramic active visual search, resulting in low object detection accuracy and poor exploration efficiency. This work proposes a global-to-local active exploration framework that, for the first time, incorporates panoramic priors to guide search and employs RoPE Rolling positional encoding to explicitly model circular topological structure. To support this approach, we introduce EAGLE-360, a large-scale, high-quality panoramic visual question answering dataset, and adopt a joint training strategy combining supervised fine-tuning (SFT) and grouped relative policy optimization (GRPO). The proposed method achieves a new state of the art on 360° visual search tasks, improving accuracy by nearly eightfold over baseline methods and significantly enhancing both exploration efficiency and robustness.

Abstract

While Multimodal Large Language Models (MLLMs) have demonstrated exceptional capabilities in standard visual understanding, adapting them for active visual search in 360$^\circ$ panoramic environments exposes fundamental limitations. Specifically, standard MLLMs struggle to effectively model inherent panoramic properties, such as severe polar distortion and continuous cylindrical topologies, which significantly degrades target detection accuracy. Consequently, existing panoramic search methods attempt to compensate by relying heavily on fragmented local viewpoints. Burdened by rigid initialization and a lack of global panoramic priors, these approaches suffer from myopic, inefficient exploration and struggle with robust error recovery when targets are out of view. To overcome these challenges, we propose EAGLE-360, a novel Embodied Active Global-to-Local Exploration framework. Rather than performing exhaustive local searches, EAGLE-360 leverages global priors to establish an initial holistic perspective, iteratively reasoning and progressively narrowing the search space. Architecturally, we adapt RoPE Rolling, a coordinate-shifting positional encoding mechanism, to seamlessly model the continuous topologies of panoramas. To facilitate this paradigm, we construct the large-scale EAGLE-360 dataset, comprising 14,000+ 4K panoramas and 70,000+ rounds of high-quality VQA dialogues. By employing a training pipeline that integrates Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO), we effectively elicit complex spatial reasoning and tool-calling capabilities. Extensive experiments demonstrate that EAGLE-360 establishes a new state-of-the-art for 360$^\circ$ visual search, achieving nearly an 8-fold increase in accuracy over the base model while significantly enhancing exploration efficiency.

Problem

• 360° visual search
• panoramic exploration
• multimodal large language models
• active vision
• spatial reasoning

Innovation

• Embodied Active Exploration
• 360-degree Panoramic Understanding
• Global-to-Local Search
• RoPE Rolling
• Multimodal LLMs