Act Wisely: Cultivating Meta-Cognitive Tool Use in Agentic Multimodal Models

TL;DR

This paper introduces HDPO, a new framework for agentic multimodal models that improves tool use efficiency and reasoning accuracy by decoupling optimization objectives, leading to fewer unnecessary tool invocations.

Contribution

It proposes HDPO, a novel approach that separates accuracy and efficiency optimization, enabling agents to better arbitrate internal knowledge and external tool use.

Findings

Metis reduces tool invocations by orders of magnitude.

Metis achieves higher reasoning accuracy.

HDPO outperforms existing reinforcement learning protocols.

Abstract

The advent of agentic multimodal models has empowered systems to actively interact with external environments. However, current agents suffer from a profound meta-cognitive deficit: they struggle to arbitrate between leveraging internal knowledge and querying external utilities. Consequently, they frequently fall prey to blind tool invocation, resorting to reflexive tool execution even when queries are resolvable from the raw visual context. This pathological behavior precipitates severe latency bottlenecks and injects extraneous noise that derails sound reasoning. Existing reinforcement learning protocols attempt to mitigate this via a scalarized reward that penalizes tool usage. Yet, this coupled formulation creates an irreconcilable optimization dilemma: an aggressive penalty suppresses essential tool use, whereas a mild penalty is entirely subsumed by the variance of the accuracy reward during advantage normalization, rendering it impotent against tool overuse. To transcend this bottleneck, we propose HDPO, a framework that reframes tool efficiency from a competing scalar objective to a strictly conditional one. By eschewing reward scalarization, HDPO maintains two orthogonal optimization channels: an accuracy channel that maximizes task correctness, and an efficiency channel that enforces execution economy exclusively within accurate trajectories via conditional advantage estimation. This decoupled architecture naturally induces a cognitive curriculum-compelling the agent to first master task resolution before refining its self-reliance. Extensive evaluations demonstrate that our resulting model, Metis, reduces tool invocations by orders of magnitude while simultaneously elevating reasoning accuracy.