Pith. sign in

REVIEW 6 cited by

Fine-Tuning Large Vision-Language Models as Decision-Making Agents via Reinforcement Learning

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2405.10292 v3 pith:EY3P7OOO submitted 2024-05-16 cs.AI cs.CLcs.CVcs.LG

classification cs.AIcs.CLcs.CVcs.LG
keywords reasoningframeworkmodelsagentsdecision-makingtaskacrossaction
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Large vision-language models (VLMs) fine-tuned on specialized visual instruction-following data have exhibited impressive language reasoning capabilities across various scenarios. However, this fine-tuning paradigm may not be able to efficiently learn optimal decision-making agents in multi-step goal-directed tasks from interactive environments. To address this challenge, we propose an algorithmic framework that fine-tunes VLMs with reinforcement learning (RL). Specifically, our framework provides a task description and then prompts the VLM to generate chain-of-thought (CoT) reasoning, enabling the VLM to efficiently explore intermediate reasoning steps that lead to the final text-based action. Next, the open-ended text output is parsed into an executable action to interact with the environment to obtain goal-directed task rewards. Finally, our framework uses these task rewards to fine-tune the entire VLM with RL. Empirically, we demonstrate that our proposed framework enhances the decision-making capabilities of VLM agents across various tasks, enabling 7b models to outperform commercial models such as GPT4-V or Gemini. Furthermore, we find that CoT reasoning is a crucial component for performance improvement, as removing the CoT reasoning results in a significant decrease in the overall performance of our method.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. OpenAlex reports about 8 citations worldwide. Full citation record

  1. PRISM: Priority-aware Rubric Internalization via Structured Multimodal Data Synthesis

    cs.LG 2026-08 conditional novelty 7.0 of 10

    PRISM trains MLLMs to act as rubric executors by synthesizing typed, prioritized rules and verification traces, lifting Qwen3-VL-4B from 9.5% to 30.1% Strict accuracy on the authors' PRISM-Eval benchmark.

  2. GUI-Libra: Training Native GUI Agents to Reason and Act with Action-aware Supervision and Partially Verifiable RL

    cs.LG 2026-02 conditional novelty 6.0 of 10

    Action-weighted SFT plus KL-regularized GRPO on a curated 81K reasoning dataset lifts open-source GUI agents by 11–22 points on online task-completion benchmarks.

  3. HALO: Human Preference Aligned Offline Reward Learning for Robot Navigation

    cs.RO 2025-08 reject novelty 6.0 of 10

    HALO learns a vision-based navigation reward from human preference rankings on egocentric video, and an IQL policy using it beats several baselines in 10-trial real-world tests.

  4. Thinking vs. Doing: Agents that Reason by Scaling Test-Time Interaction

    cs.LG 2025-06 conditional novelty 6.0 of 10

    Scaling the number of interaction steps, trained via a curriculum over rollout horizon, improves web-agent task success and outperforms scaling per-step reasoning under fixed token budgets.

  5. Qwen-Audio-VAE Technical Report

    eess.AS 2026-07 conditional novelty 5.0 of 10

    A 12.5 Hz continuous audio VAE reconstructs speech, music, and sound well while encoding 64×30s clips in 541 ms after latency-aware encoder pruning.

  6. Digi-Q: Learning Q-Value Functions for Training Device-Control Agents

    cs.LG 2025-02 conditional novelty 5.0 of 10

    An offline RL method learns a Q-function from frozen VLM features and extracts a device-control policy by imitating the best of several actions ranked by that Q-function.

Pith tools