Skip to content
Log in
Preprint

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

Changti Wu Bin Yu Zhaolong Shen Shijie Lian Xiaopeng Lin Cong Huang Zhirui Zhang Lei Zhang Kai Chen
Jul 2026 · 1 citation · 33 references
Computer Science

TL;DR

SIEVE, a structure-aware data selection method for VLA imitation learning that can surpass full-data training while using only 50% of demonstrations and 50% of training steps, suggests that reusable structure, captured through primitives and transitions, is an important signal for efficient VLA imitation learning.

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.

arXiv Semantic Scholar
View source

Similar papers

RECAP-Ψ : Advantage-Conditioned Fine-Tuning for Open-Source Humanoid VLAs

This paper applies RECAP’s three-component recipe to Ψ 0 Wei et al. (2026), a recently open-sourced humanoid VLA, providing the first reproduction of RECAP outside Physical Intelligence’s stack.

Karthik Pythireddi, Aaditya Shah, Jonathan Lu · 0 citations
Preprint Jul 2026

ExToken: Structured Exploration for Efficient Vision-Language-Action Reinforcement Fine-tuning

ExToken is introduced, a simple yet general framework that condition VLA policies on discrete behavioral priors derived from offline demonstrations for structured exploration that consistently accelerates convergence, improves task performance, and exhibits strong robustness under highly constrained interaction budgets.

Yilun Kong, Yunpeng Qing, Guozheng Ma et al. · 0 citations
Preprint Jul 2026

Simple-to-Complex Structured Demonstrations for Vision-Language-Action Learning

This work proposes a simple-to-complex structured demonstration collection strategy for VLA learning using a dual-arm robotic platform that enables VLA models to first acquire fundamental manipulation skills before learning increasingly complex task compositions, facilitating more effective learning of long-horizon manipulation tasks.

Xin Qiu, Yi Yu · 0 citations
Preprint Aug 2026

FetchMan: Learning Visual Humanoid Loco-Manipulation Policies from Simulated Experiences

Visual loco-manipulation policies that can generalize to novel scenes and objects have long been a goal of robotics research. However, today's data-hungry algorithms make collecting sufficient demonstrations a struggle for tabletop manipulation, and even more so for humanoids that must also walk and balance. Learning from simulated data and transferring that behavior to the real world, as is commonly done in locomotion, sidesteps this struggle, so we replicate that recipe for loco-manipulation. In doing so, we find that cloning synthetic demonstrations results in a low performance ceiling no matter the amount of training data. Reinforcement learning breaks through it, and refining the cloned policy with Flow-GRPO on a single sparse reward yields performance that synthetic behavior cloning cannot match. Together, these stages form our end-to-end sim-to-real pipeline spanning more than 150,000 scenes, which we use to train FetchMan. We evaluate it on FetchMan-Bench, a simulation benchmark we release, and deploy it zero-shot on a real Unitree G1, where our single-object reach-and-pick policy walks to and grasps a target across unseen scenes at 73.3% success. Finally, we extend this recipe to multi-object training, a first step toward loco-manipulation generalist policies at this data scale.

Omar Rayyan, Zhi Li, Max Argus et al. · 0 citations
Preprint Aug 2026

WorldToken: Time-First Sequence Modeling for Robotic Imitation Learning

WorldToken, a time-first policy instantiation that fuses multiview images, proprioception, and task conditioning within each policy timestep into one world token is introduced and its data-scaling and temporal-context behavior under the tested recipes are characterized.

Chunkai Yang, Andong Yang, Chao Gao · 0 citations