TACO: Tool-Augmented Credit Optimization for Agentic Tool Use

2026-06-29

AI Summary

Existing reward mechanisms struggle to accurately assess the contribution of individual tool invocations to the final answer in multimodal agents, often leading to redundant or misleading operations. To address this, this work proposes TACO, a method that enables fine-grained credit assignment through dual-channel advantage signals. TACO introduces a self-supervised, external-model-free Differential Advantage Probe Reward (DAPR) mechanism to evaluate the value of each tool call, and a parameter-agnostic Outcome-Gated Advantage Routing (OGAR) strategy to precisely propagate final rewards back to critical reasoning segments. Integrated within a GRPO reinforcement learning framework and combined with probe token insertion and a two-stage SFT+RL training pipeline, TACO significantly improves accuracy across diverse perception, reasoning, and general multimodal benchmarks while effectively reducing unnecessary tool invocations, thereby achieving on-demand tool usage.

Abstract

Agentic multimodal models perform diverse operations on an image via code and reason over the returned view, an effective paradigm for fine-grained visual question answering. However, code operations can be useful, redundant, or misleading. Outcome-only rewards cannot precisely distinguish these cases, and existing process rewards either fail to attribute final correctness to individual tool calls, or require an external judge model. To address this, we introduce Tool-Augmented Credit Optimization (TACO), a GRPO variant for code-tool agents built on two coupled advantage channels. The first, Differential Answer-Probe Reward (DAPR), is a self-supervised, judge-free tool-contribution advantage that credits each tool call by its own effect on answering correctly. Probe tokens inserted into the model's reasoning elicit its predictions with and without the tool, and the difference in outcome reward is taken as the call's value: positive for a useful call, negative for a misleading one, and zero for one that changes nothing. This reuses the existing answer checker with no auxiliary judge, and, being a difference rather than an absolute probe score, is naturally robust to probe-hacking. The second is the outcome advantage from the final answer, distributed by Outcome-Gated Advantage Routing (OGAR): a parameter-free rule that, conditioned on the call's outcome, delivers this credit only to the responsible segments, suppressing wasted tool calls without any cost term. We train TACO through a two-stage SFT+RL pipeline. Extensive experiments across perception, reasoning, and general multimodal benchmarks show that it yields consistent accuracy gains and learns to invoke its tools only when they help.

Problem

• tool use
• credit assignment
• multimodal reasoning
• reinforcement learning
• visual question answering

Innovation

• Tool-Augmented Credit Optimization
• Differential Answer-Probe Reward
• Outcome-Gated Advantage Routing
• Agentic Tool Use
• Multimodal Reasoning