arXiv AI By Zhipeng Tang, Xinda Chen, Weining Rao, Xiao Li, Wenting Tan, Yuning Wang, Xiao Shi, Xiaofang Zhao

Fewer Steps, Better Actions: Rethinking Flow-Matching Inference for VLA Policies

Read the original on arXiv AI →

The Flow has not summarised this story yet — read it at arXiv AI.

arXiv AI
3d ago

Learning Low-Latency Action Verification for Frozen Vision-Language-Action Models

The paper introduces HDH, a lightweight framework that enhances action verification for frozen vision‑language‑action (VLA) models by leveraging hindsight from the model’s own deployments. HDH operates in three stages: first, it collects autonomous rollouts from the VLA policy; second, it trains a compact verifier to predict task‑success scores for action chunks; third, it uses the verifier to select the highest‑scoring action candidate for execution. Experiments on four LIBERO‑10 tasks and three real‑world manipulation tasks show that HDH achieves an average success rate of 82.3% and improves success over the baseline policy by 22.6%.

By Wen-Han Hsieh, Ken Jiankun Zheng, Chia-Chien Chen, David M. Chan, Trevor Darrell
arXiv Machine Learning
Sep 25

Decoupled Early Exits for Task-Dependent Compute Allocation in Flow-Matching VLAs

The paper introduces a framework for Flow‑Matching Vision‑Language‑Action (VLA) models that allows independent adjustment of backbone depth, action expert depth, and denoising steps. Lightweight Exit Transformers are added at intermediate layers to enable early exits, and a KV Cache synthesis mechanism manages skipped layers so the action expert can exit deeper than the backbone. Experiments on SmolVLA and π0.5 across LIBERO and Meta‑World show that joint tuning of these compute axes reduces latency by 79.2 % and FLOPs by 31.8 %, while improving mean success rate by 5.6 %.

By Riccardo Andrea Izzo, Rimvydas Rubavicius, Gianluca Bardaro, Subramanian Ramamoorthy, Matteo Matteucci, Alessandro Suglia
arXiv Computer Vision
Oct 19

MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models

MambaAct introduces a compact Mamba-based action expert to replace the transformer-based expert in flow‑matching vision‑language‑action models, preserving the pretrained visual‑language prefix encoder while reducing the action expert size by tenfold. The new design employs a lightweight Context Compressor to embed vision‑language conditioning into each Mamba block, eliminating repeated token‑level cross‑attention. Experiments show a 64.8 % reduction in denoising latency on LIBERO and 58.8 % in real‑world deployment, with a modest 1.08 % increase in mean success rate across multiple fine‑tuning seeds.

By Khuyen Pham, Tran Nguyen Le, Edoardo Fazzari, Cesare Stefanini