SIEVE: Structure-Aware Data Selection for Imitation Learning with VLA Models

2026-07-07

AI Summary

This work addresses the inefficiency of existing data selection methods in imitation learning, which often overlook reusable behavioral structures and thus fail to leverage large-scale robotic demonstrations effectively. The authors propose SIEVE, a novel structure-aware data selection framework that decomposes demonstrations into visuomotor primitives and their transition interfaces. By performing trajectory segmentation, primitive discovery, and clustering of compositional patterns, SIEVE allocates its selection budget according to behavioral structure and optimizes structural exposure based on diminishing returns. Evaluated across multiple datasets and vision-language-action (VLA) models, SIEVE significantly outperforms current approaches, achieving superior performance using only 50% of the demonstration data and training steps compared to training on the full dataset.

Abstract

Vision-Language-Action (VLA) models are typically trained by imitation learning on large-scale robot demonstration datasets, but more data does not necessarily yield better policies due to redundancy, noise, and uneven coverage. Existing data selection methods often assess demonstrations at either the trajectory or state-action level, missing the reusable structures that compose long-horizon behaviors. In this paper, we propose SIEVE, a structure-aware data selection method for VLA imitation learning. SIEVE views demonstrations as compositions of reusable primitives and transition interfaces. It first discovers visuo-motor primitives from segmented trajectories, then allocates selection budgets to composition patterns by maximizing reuse-aware structural exposure under diminishing returns. Finally, it selects medoid trajectories within each composition-pattern bucket to retain central, stable, and imitation-friendly demonstrations. Experiments across multiple datasets, benchmarks, and VLA models show that SIEVE consistently outperforms competitive data selection baselines. Notably, SIEVE can surpass full-data training while using only 50% of demonstrations and 50% of training steps, suggesting that reusable structure, captured through primitives and transitions, is an important signal for efficient VLA imitation learning.

Problem

• imitation learning
• data selection
• Vision-Language-Action models
• reusable structures
• robot demonstration

Innovation

• structure-aware data selection
• visuo-motor primitives
• composition patterns
• reuse-aware structural exposure
• imitation learning