Hierarchical Vision Language Action Model Using Success and Failure Demonstrations

TL;DR

This paper introduces VINE, a hierarchical vision-language-action model that leverages both success and failure demonstrations to improve robustness and success rates in manipulation tasks by using failure data as a structured learning signal.

Contribution

VINE is a novel hierarchical model that incorporates failure data into planning, enabling more robust decision-making in vision-language-action tasks.

Findings

VINE improves success rates across manipulation tasks.

Failure data enhances robustness and decision-making.

Hierarchical reasoning effectively utilizes mixed-quality datasets.

Abstract

Prior Vision-Language-Action (VLA) models are typically trained on teleoperated successful demonstrations, while discarding numerous failed attempts that occur naturally during data collection. However, these failures encode where and how policies can be fragile, information that can be exploited to improve robustness. We address this problem by leveraging mixed-quality datasets to learn failure-aware reasoning at planning time. We introduce VINE, a hierarchical vision-language-action model that separates high-level reasoning (System 2) from low-level control (System 1) under a hierarchical reinforcement learning formalism, making failures usable as a structured learning signal rather than noisy supervision. System 2 performs feasibility-guided tree search over a 2D scene-graph abstraction: it proposes subgoal transitions, predicts success probabilities from both successes and failures, and prunes brittle branches before execution, effectively casting plan evaluation as feasibility scoring. The selected subgoal sequence is then passed to System 1, which executes low-level actions without modifying the agent's core skills. Trained entirely from offline teleoperation data, VINE integrates negative experience directly into the decision loop. Across challenging manipulation tasks, this approach consistently improves success rates and robustness, demonstrating that failure data is an essential resource for converting the broad competence of VLAs into robust execution.