Iterative Tool Usage Exploration for Multimodal Agents via Step-wise Preference Tuning

• URL: http://arxiv.org/abs/2504.21561v3
• Date: Tue, 20 May 2025 09:22:47 GMT
• Title: Iterative Tool Usage Exploration for Multimodal Agents via Step-wise Preference Tuning
• Authors: Pengxiang Li, Zhi Gao, Bofei Zhang, Yapeng Mi, Xiaojian Ma, Chenrui Shi, Tao Yuan, Yuwei Wu, Yunde Jia, Song-Chun Zhu, Qing Li,
• Abstract summary: Multimodal agents, which integrate a controller e.g., a vision language model, with external tools, have demonstrated remarkable capabilities in tackling complex multimodal tasks.<n>Existing approaches for training these agents depend on extensive human-annotated task-answer pairs and tool trajectories.<n>We propose an iterative tool usage exploration method for multimodal agents without any pre-collected data, namely SPORT.<n>SPORT has four iterative components: task synthesis, step sampling, step verification, and preference tuning.
• Score: 69.32855772335624
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multimodal agents, which integrate a controller e.g., a vision language model) with external tools, have demonstrated remarkable capabilities in tackling complex multimodal tasks. Existing approaches for training these agents, both supervised fine-tuning and reinforcement learning, depend on extensive human-annotated task-answer pairs and tool trajectories. However, for complex multimodal tasks, such annotations are prohibitively expensive or impractical to obtain. In this paper, we propose an iterative tool usage exploration method for multimodal agents without any pre-collected data, namely SPORT, via step-wise preference optimization to refine the trajectories of tool usage. Our method enables multimodal agents to autonomously discover effective tool usage strategies through self-exploration and optimization, eliminating the bottleneck of human annotation. SPORT has four iterative components: task synthesis, step sampling, step verification, and preference tuning. We first synthesize multimodal tasks using language models. Then, we introduce a novel trajectory exploration scheme, where step sampling and step verification are executed alternately to solve synthesized tasks. In step sampling, the agent tries different tools and obtains corresponding results. In step verification, we employ a verifier to provide AI feedback to construct step-wise preference data. The data is subsequently used to update the controller for tool usage through preference tuning, producing a SPORT agent. By interacting with real environments, the SPORT agent gradually evolves into a more refined and capable system. Evaluation in the GTA and GAIA benchmarks shows that the SPORT agent achieves 6.41% and 3.64% improvements, underscoring the generalization and effectiveness introduced by our method. The project page is https://SPORT-Agents.github.io.

Related papers

• LAM SIMULATOR: Advancing Data Generation for Large Action Model Training via Online Exploration and Trajectory Feedback [121.78866929908871]
  Large Action Models (LAMs) for AI Agents offer incredible potential but face challenges due to the need for high-quality training data.<n>We present LAM SIMULATOR, a comprehensive framework designed for online exploration of agentic tasks with high-quality feedback.<n>Our framework features a dynamic task query generator, an extensive collection of tools, and an interactive environment where Large Language Model (LLM) Agents can call tools and receive real-time feedback.
  arXiv  Detail & Related papers  (2025-06-02T22:36:02Z)
• LaMDAgent: An Autonomous Framework for Post-Training Pipeline Optimization via LLM Agents [3.6117068575553595]
  We introduce LaMDAgent, a framework that autonomously constructs and optimize full post-training pipelines.<n>LaMDAgent improves tool-use accuracy by 9.0 points while preserving instruction-following capabilities.<n>It uncovers effective post-training strategies that are often overlooked by conventional human-driven exploration.
  arXiv  Detail & Related papers  (2025-05-28T04:30:51Z)
• T^2Agent A Tool-augmented Multimodal Misinformation Detection Agent with Monte Carlo Tree Search [51.91311158085973]
  multimodal misinformation often arises from mixed forgery sources, requiring dynamic reasoning and adaptive verification.<n>We propose T2Agent, a novel misinformation detection agent that incorporates a toolkit with Monte Carlo Tree Search.<n>Extensive experiments show that T2Agent consistently outperforms existing baselines on challenging mixed-source multimodal misinformation benchmarks.
  arXiv  Detail & Related papers  (2025-05-26T09:50:55Z)
• ToolACE-R: Tool Learning with Adaptive Self-Refinement [84.69651852838794]
  Tool learning allows Large Language Models to leverage external tools for solving complex user tasks.<n>We propose ToolACE-R, a novel method that introduces adaptive self-refinement for tool invocations.<n>Our results demonstrate the effectiveness of the proposed method, which is compatible with base models of various sizes.
  arXiv  Detail & Related papers  (2025-04-02T06:38:56Z)
• STEVE: A Step Verification Pipeline for Computer-use Agent Training [84.24814828303163]
  STEVE is a step verification pipeline for computer-use agent training.<n> GPT-4o is used to verify the correctness of each step in the trajectories based on the screens before and after the action execution.<n>Our agent outperforms supervised finetuning by leveraging both positive and negative actions within a trajectory.
  arXiv  Detail & Related papers  (2025-03-16T14:53:43Z)
• Multi-Agent Geospatial Copilots for Remote Sensing Workflows [1.8241060496411214]
  GeoLLM-Squad introduces the novel multi-agent paradigm to remote sensing (RS)<n>Unlike existing single-agent approaches that rely on monolithic large language models (LLM), GeoLLM-Squad separates agentic orchestration from geospatial task-solving.<n>Our work enables the modular integration of diverse applications, spanning urban monitoring, forestry protection, climate analysis, and agriculture studies.
  arXiv  Detail & Related papers  (2025-01-27T17:54:31Z)
• Pilot: Building the Federated Multimodal Instruction Tuning Framework [79.56362403673354]
  Our framework integrates two stages of "adapter on adapter" into the connector of the vision encoder and the LLM. In stage 1, we extract task-specific features and client-specific features from visual information. In stage 2, we build the cross-task Mixture-of-Adapters(CT-MoA) module to perform cross-task interaction.
  arXiv  Detail & Related papers  (2025-01-23T07:49:24Z)
• Multi-Agent Sampling: Scaling Inference Compute for Data Synthesis with Tree Search-Based Agentic Collaboration [81.45763823762682]
  This work aims to bridge the gap by investigating the problem of data synthesis through multi-agent sampling.<n>We introduce Tree Search-based Orchestrated Agents(TOA), where the workflow evolves iteratively during the sequential sampling process.<n>Our experiments on alignment, machine translation, and mathematical reasoning demonstrate that multi-agent sampling significantly outperforms single-agent sampling as inference compute scales.
  arXiv  Detail & Related papers  (2024-12-22T15:16:44Z)
• TrajAgent: An Agent Framework for Unified Trajectory Modelling [7.007450097312181]
  Trajectory modeling has widespread applications in areas such as life services, urban transportation, and public administration. We propose TrajAgent, a large language model-based agentic framework, to unify various trajectory modelling tasks.
  arXiv  Detail & Related papers  (2024-10-27T13:51:09Z)
• ComfyBench: Benchmarking LLM-based Agents in ComfyUI for Autonomously Designing Collaborative AI Systems [80.69865295743149]
  This work attempts to study using LLM-based agents to design collaborative AI systems autonomously.<n>Based on ComfyBench, we develop ComfyAgent, a framework that empowers agents to autonomously design collaborative AI systems by generating.<n>While ComfyAgent achieves a comparable resolve rate to o1-preview and significantly surpasses other agents on ComfyBench, ComfyAgent has resolved only 15% of creative tasks.
  arXiv  Detail & Related papers  (2024-09-02T17:44:10Z)
• Re-Invoke: Tool Invocation Rewriting for Zero-Shot Tool Retrieval [47.81307125613145]
  Re-Invoke is an unsupervised tool retrieval method designed to scale effectively to large toolsets without training. We employ a novel multi-view similarity ranking strategy based on intents to pinpoint the most relevant tools for each query. Our evaluation demonstrates that Re-Invoke significantly outperforms state-of-the-art alternatives in both single-tool and multi-tool scenarios.
  arXiv  Detail & Related papers  (2024-08-03T22:49:27Z)
• Octo-planner: On-device Language Model for Planner-Action Agents [19.627197141903505]
  Planner-Action framework separates planning and action execution into two distinct components. Agent first responds to user queries by decomposing tasks into a sequence of sub-steps, which are then executed by the action agent. We employ model fine-tuning instead of in-context learning, reducing computational costs and energy consumption.
  arXiv  Detail & Related papers  (2024-06-26T05:40:10Z)
• MLAgentBench: Evaluating Language Agents on Machine Learning Experimentation [96.71370747681078]
  We introduce MLAgentBench, a suite of 13 tasks ranging from improving model performance on CIFAR-10 to recent research problems like BabyLM. For each task, an agent can perform actions like reading/writing files, executing code, and inspecting outputs. We benchmark agents based on Claude v1.0, Claude v2.1, Claude v3 Opus, GPT-4, GPT-4-turbo, Gemini-Pro, and Mixtral and find that a Claude v3 Opus agent is the best in terms of success rate.
  arXiv  Detail & Related papers  (2023-10-05T04:06:12Z)
• Learning Generalizable Tool-use Skills through Trajectory Generation [13.879860388944214]
  We train a single model on four different deformable object manipulation tasks. The model generalizes to various novel tools, significantly outperforming baselines. We further test our trained policy in the real world with unseen tools, where it achieves the performance comparable to human.
  arXiv  Detail & Related papers  (2023-09-29T21:32:42Z)
• AutoAgents: A Framework for Automatic Agent Generation [27.74332323317923]
  AutoAgents is an innovative framework that adaptively generates and coordinates multiple specialized agents to build an AI team according to different tasks. Our experiments on various benchmarks demonstrate that AutoAgents generates more coherent and accurate solutions than the existing multi-agent methods.
  arXiv  Detail & Related papers  (2023-09-29T14:46:30Z)
• Multitask Adaptation by Retrospective Exploration with Learned World Models [77.34726150561087]
  We propose a meta-learned addressing model called RAMa that provides training samples for the MBRL agent taken from task-agnostic storage. The model is trained to maximize the expected agent's performance by selecting promising trajectories solving prior tasks from the storage.
  arXiv  Detail & Related papers  (2021-10-25T20:02:57Z)
• Meta Reinforcement Learning with Autonomous Inference of Subtask Dependencies [57.27944046925876]
  We propose and address a novel few-shot RL problem, where a task is characterized by a subtask graph. Instead of directly learning a meta-policy, we develop a Meta-learner with Subtask Graph Inference. Our experiment results on two grid-world domains and StarCraft II environments show that the proposed method is able to accurately infer the latent task parameter.
  arXiv  Detail & Related papers  (2020-01-01T17:34:00Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.