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

Published July 8, 2026 · Category: Robotics

Overview

arXiv:2607.06442v1 Announce Type: new 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.

Source

Originally published at arxiv.org.

Related Articles

• Prior-First, Condition-Second: Scalable and Controllable Hand Motion Completion
• Uncertainty-Aware Velocity Correction for Proprioceptive Vehicle Localization using Evidential Mamba
• IMR: Iterative Mode-World Weighted Regression for Multi-Agent Trajectory Prediction