AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Reflex: Real-Time Vision-Language-Action Control through Streaming Inference

Yuanchun Guo, Bingyan Liu

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Block-sparse attention kernels

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Partitioned Attention (cache static/sliding/dynamic regions) (ablation)Reflex on Pi0.5 · LIBEROinference latency (ms): 135.2 -> 61.5Improved
add async execution and future-conditional overlap scheduling (ablation)Reflex on Pi0.5 · LIBEROreaction latency (ms): 151.9 -> 82.5Improved
use timestep-aware partitioned caching instead of naive KV caching (ablation)Flow matching VLA · LIBEROsuccess rate (%): 12.5 -> 85.4Improved
use Partitioned Attention instead of full recomputation (ablation)Flow matching VLA · LIBEROsuccess rate (%): 85.2 -> 85.4No clear effect
add AdaRMSNorm to BF16 inference (ablation)Reflex · BF16 stress testmax stable steps: 120-220 -> >2000Improved
use AdaRMSNorm instead of FP32 norm-only (ablation)Reflex · BF16 stress testadded latency (ms): +0.8 -> +0.4Improved
add fused FlashNorm and FusedAdaLN kernels (ablation)Reflex action expert · Action expert latencyinference latency (ms): 22.8 -> 18.6Improved
add future-state predictor at 4 delay steps (ablation)Reflex · Kinetix controlled delaysuccess rate (%): +11pp over removing the predictorImproved
increase context window K from 10 to 50 (ablation)Reflex · LIBEROspeedup: 2.58x -> below 1xHurt
use K=1 context window (ablation)Reflex · LIBEROaccuracyHurt
use Reflex instead of synchronous inference on dynamic KinetixPi0.5 · Kinetixsuccess rate: +7.4%Improved
use Reflex instead of synchronous inference on LIBERO-LongPi0.5 · LIBERO-Longsuccess rate (%): 68.8 -> 72.4Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy