Newest papers

FIT-MAC: A Closed-Loop Multi-Agent Framework for Robust Vision-Language-Action Models

Jiatong Li, Jisheng Dang, Qiangqiang Ma, Bimei Wang, Jingwen Zhao, Wenjun Ma, Juncheng Li, Hong Peng, Bin Hu · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28109 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Os modelos de visão de linguagem-ação (VLA) têm alcançado um desempenho forte sobre os benchmarks de manipulação robótica padrão. No entanto, sua confiabilidade pode diminuir substancialmente quando as condições de tarefa se desviam da configuração nominal. As perturbações nas posições de objeto e as instruções de tarefa podem levar à falha de tarefa completa. As políticas de VLA existentes geralmente carecem de um mecanismo de tempo de execução explícito para distinguir esses modos de falha e selecionar uma estratégia de recuperação apropriada. Para mitigar sistematicamente este problema, propomos o FIT-MAC, um framework de controle de múltiplos agentes conscientes de falhas e de circuito fechado para melhorar a robustez das políticas de VLA na inferência. O FIT

Machine translation (Meta M2M100); the English original follows.

Abstract

Vision-language-action (VLA) models have achieved strong performance on standard robotic manipulation benchmarks. However, their reliability can degrade substantially when task conditions deviate from the nominal setting. Perturbations in object positions and task instructions may lead to complete task failure. Existing VLA policies generally lack an explicit runtime mechanism for distinguishing these failure modes and selecting an appropriate recovery strategy. To systematically mitigate this issue, we propose FIT-MAC, a failure-aware, closed-loop multi-agent control framework for improving the robustness of VLA policies at inference time. FIT-MAC augments a VLA policy with an inference-time control layer. First, the Planner Agent parses the language instruction into a structured task specification that captures the manipulation goal and spatial constraints, and the Grounder Agent resolves each instruction-level reference to a visual target. Next, the Verifier Agent tracks phase-specific execution progress, distinguishes target-binding errors from local execution failures, and determines whether intervention is warranted. The Correction Agent then routes permitted interventions to target rebinding or bounded grasping or placement recovery, depending on the failure type. When no intervention is required, the pretrained VLA remains the default action generator. Extensive experiments on LIBERO-PRO demonstrate that FIT-MAC improves robustness across multiple VLA backbones, consistently increasing success rates and achieving state-of-the-art performance under diverse perturbations.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Robustness (evolution), Planner, Inference, Task (project management), Artificial intelligence

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange