VideoSearcher: Empowering Video Deep Research with Multi-Tool Agentic Reasoning via Reinforcement Learning

2026-07-02

AI Summary

This work addresses the limitations of existing video understanding approaches, which are constrained by closed-world contexts and lack the capacity to explore multimodal evidence in open-world settings. Current benchmarks predominantly rely on text-only retrieval, neglecting crucial visual information. To overcome these challenges, we propose VideoSearcher, a closed-loop, multi-tool agent framework that performs joint reasoning through temporal localization, spatial focusing, and multimodal retrieval, progressively anchoring visual cues, retrieving evidence, and generating answers along its reasoning trajectory. We introduce Bi-branch Sequence Policy Optimization (BiSPO), a reinforcement learning algorithm that decouples tool invocation from answer generation optimization, and construct VideoSearch-QA—the first benchmark tailored for open-world video grounding and multimodal search-based reasoning. Experiments demonstrate that our approach significantly outperforms existing open-source agent baselines across multiple search-oriented and multimodal comprehension tasks.

Abstract

Video understanding is moving beyond closed-context perception toward open-world evidence exploration, a paradigm formalized as Video Deep Research (VDR). However, existing multimodal search agents primarily target static images, and the current VDR benchmark relies on text-centric retrieval that discards crucial visual information. To address these limitations, we propose VideoSearcher, a closed-loop agentic framework that empowers Vision-Language Models with multi-tool reasoning for VDR. VideoSearcher unifies temporal localization, spatial focusing, and multimodal search within a single reasoning trajectory, enabling agents to progressively ground visual clues, retrieve relevant evidence, and synthesize answers. To optimize knowledge-intensive reasoning trajectories, we propose Bi-branch Sequence Policy Optimization (BiSPO), a reinforcement learning algorithm that decouples tool-invocation optimization from answer-accuracy optimization. This design provides stable learning signals for both evidence-grounded reasoning and purposeful tool use. Furthermore, we construct VideoSearch-QA, the first benchmark designed to evaluate open-world video information grounding and multimodal search-based reasoning. Extensive experiments demonstrate that VideoSearcher significantly outperforms prior open-source agentic baselines across various search-oriented and multimodal understanding benchmarks.

Problem

• Video Deep Research
• multimodal search
• open-world video understanding
• visual information grounding
• video reasoning

Innovation

• Video Deep Research
• Multi-Tool Agentic Reasoning
• Reinforcement Learning
• Vision-Language Models
• Multimodal Search