MATRIX: Multimodal Agent Tuning for Robust Tool-Use Reasoning
This addresses the problem of scalable and effective multimodal tool-use reasoning for AI agents, though it is incremental as it builds on existing VLM and tuning methods.
The paper tackles the challenge of limited high-quality multimodal trajectories for vision language models (VLMs) used as controllers with external tools, by introducing a framework that automatically synthesizes such trajectories and trains a VLM controller, resulting in consistent performance improvements across three benchmarks.
Vision language models (VLMs) are increasingly deployed as controllers with access to external tools for complex reasoning and decision-making, yet their effectiveness remains limited by the scarcity of high-quality multimodal trajectories and the cost of manual annotation. We address this challenge with a vision-centric agent tuning framework that automatically synthesizes multimodal trajectories, generates step-wise preference pairs, and trains a VLM controller for robust tool-use reasoning. Our pipeline first constructs M-TRACE, a large-scale dataset of 28.5K multimodal tasks with 177K verified trajectories, enabling imitation-based trajectory tuning. Building on this, we develop MATRIX Agent, a controller finetuned on M-TRACE for step-wise tool reasoning. To achieve finer alignment, we further introduce Pref-X, a set of 11K automatically generated preference pairs, and optimize MATRIX on it via step-wise preference learning. Across three benchmarks, Agent-X, GTA, and GAIA, MATRIX consistently surpasses both open- and closed-source VLMs, demonstrating scalable and effective multimodal tool use. Our data and code is avaliable at https://github.com/mbzuai-oryx/MATRIX.