XS-VLA: Coupling Coarse-grained Spatial Distillation with Latent Flow Matching for Lightweight Robotic Control

2026-07-05

AI Summary

This work addresses the limitations of lightweight vision-language-action models in robotic control, which often suffer from weak spatial perception and suboptimal policy performance, struggling to balance real-time inference with robust spatial understanding. The authors propose a two-stage framework: first, coarse-grained spatial semantic distillation transfers spatial knowledge from a large model to a compact SmolVLM2-0.25B backbone; second, a conditional variational autoencoder (CVAE) combined with latent flow matching yields an efficient latent flow policy. This approach uniquely integrates spatial knowledge distillation with flow matching, substantially enhancing the small model’s spatial reasoning and multimodal action generation capabilities. Evaluated on the LIBERO benchmark, the method achieves state-of-the-art performance under 0.5B parameters, improving average task success by 7.2% (notably 23% on LIBERO-Long) while running 3.2× faster than existing lightweight policies.

Abstract

Large Vision-Language Models (LVLMs) have shown strong multimodal understanding and spatial grounding, but their computational cost limits real-time robotic control. In contrast, lightweight models are suitable for edge deployment but often suffer from "spatial blindness", namely weak native spatial prediction ability. Training Vision-Language-Action (VLA) models on mixed human demonstrations can also degrade policy performance due to highly diverse behaviors. To address these limitations, we propose XS-VLA, a two-stage framework for efficient and spatially grounded robotic manipulation. First, we distill spatial semantic knowledge from Qwen3-VL-4B into the SmolVLM2-0.25B backbone by fine-tuning on curated coarse-grained spatial descriptions, turning the lightweight model into a spatially grounded engine. Second, we use this enhanced backbone to condition a Latent Flow Matching policy. Unlike deterministic controllers, our policy combines a Conditional Variational Autoencoder (CVAE) with Flow Matching dynamics to model complex multimodal action distributions. On the LIBERO benchmark, XS-VLA achieves state-of-the-art performance among models with fewer than 0.5B parameters. It improves average success rates by up to 7.2 percent, including a 23 percent gain on LIBERO-Long, over the SmolVLA 0.25B baseline, and outperforms the larger 2.2B vanilla SmolVLA. Ablations show that spatial tuning and generative latent flow control substantially improve lightweight VLA performance, delivering a 3.2 times speedup in mission execution over the previous lightweight flow matching policy.

Problem

• spatial blindness
• lightweight robotic control
• Vision-Language-Action models
• multimodal action distributions
• real-time deployment

Innovation

• Spatial Distillation
• Latent Flow Matching
• Vision-Language-Action
• Lightweight Robotic Control
• Coarse-grained Spatial Grounding