Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SimVLA: A Simple VLA Baseline for Robotic Manipulation

Yuankai Luo, Woping Chen, Tong Liang, Baiqiao Wang, Zhenguo Li

TL;DR

SimVLA is introduced, a streamlined baseline designed to establish a robust, reproducible baseline that enables clear attribution of empirical gains to future architectural innovations and demonstrates that a minimal design can achieve state-of-the-art performance.

Abstract

Vision-Language-Action (VLA) models have emerged as a promising paradigm for general-purpose robotic manipulation, leveraging large-scale pre-training to achieve strong performance. The field has rapidly evolved with additional spatial priors and diverse architectural innovations. However, these advancements are often accompanied by varying training recipes and implementation details, which can make it challenging to disentangle the precise source of empirical gains. In this work, we introduce SimVLA, a streamlined baseline designed to establish a transparent reference point for VLA research. By strictly decoupling perception from control, using a standard vision-language backbone and a lightweight action head, and standardizing critical training dynamics, we demonstrate that a minimal design can achieve state-of-the-art performance. Despite having only 0.5B parameters, SimVLA outperforms multi-billion-parameter models on standard simulation benchmarks without robot pretraining. SimVLA also reaches on-par real-robot performance compared to pi0.5. Our results establish SimVLA as a robust, reproducible baseline that enables clear attribution of empirical gains to future architectural innovations. Website: https://frontierrobo.github.io/SimVLA

SimVLA: A Simple VLA Baseline for Robotic Manipulation

TL;DR

SimVLA is introduced, a streamlined baseline designed to establish a robust, reproducible baseline that enables clear attribution of empirical gains to future architectural innovations and demonstrates that a minimal design can achieve state-of-the-art performance.

Abstract

Vision-Language-Action (VLA) models have emerged as a promising paradigm for general-purpose robotic manipulation, leveraging large-scale pre-training to achieve strong performance. The field has rapidly evolved with additional spatial priors and diverse architectural innovations. However, these advancements are often accompanied by varying training recipes and implementation details, which can make it challenging to disentangle the precise source of empirical gains. In this work, we introduce SimVLA, a streamlined baseline designed to establish a transparent reference point for VLA research. By strictly decoupling perception from control, using a standard vision-language backbone and a lightweight action head, and standardizing critical training dynamics, we demonstrate that a minimal design can achieve state-of-the-art performance. Despite having only 0.5B parameters, SimVLA outperforms multi-billion-parameter models on standard simulation benchmarks without robot pretraining. SimVLA also reaches on-par real-robot performance compared to pi0.5. Our results establish SimVLA as a robust, reproducible baseline that enables clear attribution of empirical gains to future architectural innovations. Website: https://frontierrobo.github.io/SimVLA
Paper Structure (29 sections, 2 equations, 3 figures, 8 tables)

This paper contains 29 sections, 2 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. SimVLA: A Simple VLA Baseline
  4. Preliminaries
  5. SimVLA Architecture
  6. Training and Inference Recipe
  7. Evaluation
  8. Experimental Setup
  9. Simulation Benchmarks
  10. Real-robot (Galaxea R1 Lite)
  11. Simulation Results
  12. Ablation Studies
  13. Real-robot Results
  14. Conclusion
  15. Datasets and Experimental Details
  16. ...and 14 more sections

Figures (3)

  • Figure 1: Out-of-box real-robot task examples. We deploy SimVLA without any additional fine-tuning on our held-out scenes and evaluate it on a set of multi-stage tasks that require both dexterous manipulation and semantic understanding.
  • Figure 2: SimVLA overview. SimVLA is a minimal baseline: a VLM encoder produces fused vision-language tokens once per control step, and a lightweight action transformer performs flow-matching denoising to generate a continuous action chunk.
  • Figure 3: Real-robot zero-shot results on Galaxea R1 Lite.