VLANeXt: Recipes for Building Strong VLA Models

TL;DR

VLANeXt addresses the fragmented VLA landscape by introducing a unified design-space framework and a practical recipe for building strong VLA models. Through systematic ablations across foundational components, perception, and action modeling, it derives 12 actionable findings and demonstrates that a compact model can achieve state-of-the-art results on LIBERO and LIBERO-plus, with robust real-world performance. The approach emphasizes soft VLM–policy coupling, VLM-side proprioception conditioning, longer action chunks, multi-view perception, and a frequency-domain auxiliary loss, while balancing computational cost against gains. A lightweight, reproducible codebase is released to standardize training/evaluation and stimulate further progress in VLA research.

Abstract

Following the rise of large foundation models, Vision-Language-Action models (VLAs) emerged, leveraging strong visual and language understanding for general-purpose policy learning. Yet, the current VLA landscape remains fragmented and exploratory. Although many groups have proposed their own VLA models, inconsistencies in training protocols and evaluation settings make it difficult to identify which design choices truly matter. To bring structure to this evolving space, we reexamine the VLA design space under a unified framework and evaluation setup. Starting from a simple VLA baseline similar to RT-2 and OpenVLA, we systematically dissect design choices along three dimensions: foundational components, perception essentials, and action modelling perspectives. From this study, we distill 12 key findings that together form a practical recipe for building strong VLA models. The outcome of this exploration is a simple yet effective model, VLANeXt. VLANeXt outperforms prior state-of-the-art methods on the LIBERO and LIBERO-plus benchmarks and demonstrates strong generalization in real-world experiments. We will release a unified, easy-to-use codebase that serves as a common platform for the community to reproduce our findings, explore the design space, and build new VLA variants on top of a shared foundation.

Figures (11)

• Figure 1: Performance comparison on the LIBERO and LIBERO-plus benchmarks. We compare VLANeXt with representative VLA baselines across model scales. Despite its smaller model size, VLANeXt achieves higher success rates than prior methods on both standard task performance (LIBERO) and robustness/generalization (LIBERO-plus), demonstrating the effectiveness of the design recipe distilled in this work.
• Figure 2: Ablation trajectory across the VLA design space (spatial suite). We progressively evolve a baseline VLA through changes in foundational components, perception, and action modeling. Results are reported on LIBERO initially, and on LIBERO-plus once LIBERO performance saturates, providing a more sensitive test of robustness and generalization. The trajectory culminates in the final VLANeXt model (2.5B) vs. OpenVLA-OFT (7B).
• Figure 3: Design choices for the policy module.
• Figure 4: Design choices for the VLM-Policy connection.
• Figure 5: Design choices for proprioception conditioning.