Search papers, labs, and topics across Lattice.
This paper introduces Reflex, a framework that enables real-time streaming inference for flow matching Vision-Language-Action (VLA) models by leveraging the Timestep-Invariance Property to partition attention context. By implementing $O(1)$ incremental cache updates and introducing an adaptive normalization layer called AdaRMSNorm, Reflex achieves significant improvements in inference speed and stability. The framework demonstrates a 2.58脳 speedup and 50Hz stable streaming on benchmarks, reducing reaction latency by up to 54%, making it suitable for practical robotic applications.
Reflex achieves a 2.58脳 inference speedup while maintaining high-frequency stability, revolutionizing real-time control in robotics.
Flow matching Vision-Language-Action (VLA) models promise precise continuous control, but their iterative denoising nature introduces fundamental incompatibilities with real-time robotics: global timestep injection invalidates KV-caching, forcing a choice between slow $O(N^2)$ re-computation or mathematically incorrect cache reuse. We present \textbf{Reflex}, a framework that enables \textit{real-time streaming inference} for flow matching policies by exploiting the \textit{Timestep-Invariance Property} -- that perception encoders are functionally independent of the denoising loop. Reflex partitions the attention context into static, sliding, and dynamic regions, enabling $O(1)$ incremental cache updates while preserving full-batch-equivalent attention outputs for fixed inputs. To ensure stability under continuous high-frequency inference, we introduce \textit{AdaRMSNorm}, an adaptive normalization layer that prevents BFloat16 numerical collapse by gating on flow phase. We further maximize throughput through an \textit{async pipeline} that decouples visual encoding from action generation, combined with \textit{operator fusion} that reduces kernel overhead. On LIBERO and Kinetix benchmarks, Reflex achieves a 2.58$\times$ inference speedup and 50Hz stable streaming, reducing reaction latency by up to 54\% and enabling efficient deployment without performance degradation.
