CycleVLA: Proactive Self-Correcting Vision-Language-Action Models via Subtask Backtracking and Minimum Bayes Risk Decoding

2026-01-05

arXiv.org

AI Summary

This work proposes a novel Vision-Language-Action (VLA) framework endowed with proactive self-correction capabilities to address the limitation of existing robot failure detection methods, which typically rely on post-hoc interventions and cannot prevent errors during execution. The framework predicts potential failures at critical subtask transition points and triggers a backtracking mechanism, augmented by Minimum Bayes Risk (MBR) decoding to enhance retry success rates. It represents the first approach enabling VLA models to actively anticipate failures and backtrack to prior subtasks during execution, further introducing an MBR-based zero-shot test-time scaling strategy. Experimental results demonstrate that the proposed framework significantly improves task success rates across both well-trained and under-trained VLA models, confirming its effectiveness and robustness.

Technology Category

• Computer Vision: Language and Vision
• Intelligent Robots: Learning & Optimization for ROB
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Current work on robot failure detection and correction typically operate in a post hoc manner, analyzing errors and applying corrections only after failures occur. This work introduces CycleVLA, a system that equips Vision-Language-Action models (VLAs) with proactive self-correction, the capability to anticipate incipient failures and recover before they fully manifest during execution. CycleVLA achieves this by integrating a progress-aware VLA that flags critical subtask transition points where failures most frequently occur, a VLM-based failure predictor and planner that triggers subtask backtracking upon predicted failure, and a test-time scaling strategy based on Minimum Bayes Risk (MBR) decoding to improve retry success after backtracking. Extensive experiments show that CycleVLA improves performance for both well-trained and under-trained VLAs, and that MBR serves as an effective zero-shot test-time scaling strategy for VLAs. Project Page: https://dannymcy.github.io/cyclevla/

Problem

• failure detection
• self-correction
• proactive recovery
• vision-language-action models
• robot execution

Innovation

• proactive self-correction
• subtask backtracking
• Minimum Bayes Risk decoding
• Vision-Language-Action models
• failure prediction