Vision-Language-Action (VLA) models are increasingly promising for robotic manipulation, yet their real-world deployment remains bottlenecked by high inference latency and unstable asynchronous execution. This challenge is particularly pronounced in flow-matching-based VLA models, where action decoding requires multiple iterative steps conditioned on the VLM context. While efficient inference methods improve control frequency and asynchronous methods reduce execution idle time, existing approaches often fail to jointly achieve low-latency inference and accurate, temporally consistent asynchronous execution. We introduce FlashVLA, a streaming action decoding framework that addresses both challenges in a unified formulation. FlashVLA maintains a streaming action buffer with multiple chunks at different noise levels and decodes them using chunk-wise causal attention. This design allows FlashVLA to produce one executable action chunk per inference step. Moreover, its chunk-wise autoregressive formulation implicitly preserves action continuity, enabling smooth asynchronous execution without extra future-state conditioning. Across extensive simulated and real-world experiments, FlashVLA substantially improves inference speed while maintaining strong task performance. It can achieve ≥30,Hz control frequency on a single GPU with smooth asynchronous inference in real-world deployment.
Figures & tables
Figure 1: (a) Two prevailing VLA inference patterns and FlashVLA’s joint solution. (b) Per-policy-invocation profile of π0.5 and FlashVLA. Action decoding is the dominant bottleneck.
Figure 2: Overview of FlashVLA. Left : Multi-buffer joint fine-tuning packs all cold-start configurations for one observation into a single sample. Attention masks isolate buffers, and chunk-wise causal attention lets noisier chunks attend only to cleaner ones. Right : Streaming inference runs the buffer as a queue: cold-start fills it with N−1 warm-up passes, while steady streaming advances all chunks, executes the cleanest one, and appends fresh noise. Gray squares denote padding.
Method
Success Rate (%)
Efficiency
Spatial
Object
Goal
Long
Average
Steps
Time (s)
Time / Step (ms)
π0.5
98.8
98.2
98.0
92.4
96.9
156.0
8.4
53.8
+ VLASH (d=1)
98.8
99.2
96.7
94.4
97.2 ( ↑ 0.3 )
153.9
7.2
46.8 ( 1.15 × )
+ VLASH (d=4)
92.5
96.9
93.3
89.6
93.1 ( ↓ 3.8 )
176.7
5.8
32.8 ( 1.64 × )
+ StreamingVLA (d=1)
96.6
96.6
95.4
91.0
94.9 ( ↓ 2.0 )
n/a
n/a
31.6 ( 1.70 × )
+ FlashVLA (d=1)
98.8
99.6
97.6
95.4
97.8 ( ↑ 0.9 )
158.0
3.5
22.1 ( 2.43 × )
Table 1: Asynchronous performance on LIBERO at one-step delay ( d=1 ) unless noted. FlashVLA reaches the largest speedup over π0.5 (2.43 × ), while improving average success rate from 96.9% to 97.8%. Other asynchronous baselines either provide smaller speedups or lose success rate. Steps : actions executed per episode (averaged over 2,000 episodes); Time (s) : per-episode completion time.
Figure 3: Success rate and time per step across asynchronous delay d on LIBERO and RoboTwin 2.0. Both VLASH and FlashVLA are built on the same π0.5 baseline. FlashVLA remains above the synchronous success rate across d=1 – 4 while achieving lower time per step. Success rate is averaged over four LIBERO suites and the clean and randomized evaluations of 50 RoboTwin 2.0 tasks; dashed lines denote synchronous π0.5 .
Figure 5
Method
LIBERO
RoboTwin 2.0
Spatial
Object
Goal
Long
Avg.
Clean
Random
Avg.
π0.5
98.8
98.2
98.0
92.4
96.9
86.1
85.8
86.0
+FlashVLA
98.6
99.0
97.8
96.2
97.9 ( ↑ 1.0)
90.8
90.2
90.5 ( ↑ 4.5)
Table 3: Synchronous success rates on LIBERO and RoboTwin 2.0. On LIBERO, FlashVLA improves π0.5 by 1.0 points on average, with a 3.8-point gain on LIBERO-Long. RoboTwin 2.0 reports 50-task multitask evaluation at d=0 , improving the average across clean and randomized settings from 86.0% to 90.5%.
Method
Short Horizon
Medium Horizon
Long Horizon
Clean
Random
Avg.
Clean
Random
Avg.
Clean
Random
Avg.
π0.5
93.5
94.2
93.9
81.9
80.4
81.1
54.2
51.8
53.0
+FlashVLA
93.2
93.3
93.2 ( ↓ 0.7 )
86.0
84.8
85.4 ( ↑ 4.3 )
90.8
88.4
89.6 ( ↑ 36.6 )
Table 4: Synchronous performance on all 50 RoboTwin 2.0 tasks, grouped by task horizon. FlashVLA preserves short-horizon performance and improves average success by 4.3 points on medium-horizon tasks and 36.6 points on long-horizon tasks.
Method
Avg. SR (%)
Time / Step (ms)
Inference Latency (ms)
LIBERO
SmolVLA
80.1
41.2
19.7
+ FlashVLA ( d=0 )
80.1 ( ↑ 0.0 )
29.7 ( 1.39 × )
10.1 ( 1.95 × )
+ FlashVLA ( d=1 )
79.5 ( ↓ 0.6 )
28.7 ( 1.44 × )
–
RoboTwin 2.0
LingBot-VLA
85.2
56.7
70.6
Table 5: Cross-architecture generalization across SmolVLA and LingBot-VLA. With no or only minimal additional parameters, FlashVLA preserves SmolVLA success and improves LingBot-VLA by 3.4–4.1 points while reducing inference latency by 1.95 × –2.81 × . Here, d=0 denotes synchronous execution and d=1 denotes one-step asynchronous execution; delay-independent inference latency is reported only at d=0 .
Figure 5: Real-world tasks and results. Top: the short-, medium-, and long-horizon tasks used for evaluation. Bottom: task score and average completion time over successful trials. FlashVLA outperforms all three π0.5 configurations on every task and averages a 1.3 × completion-time speedup over synchronous execution.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Task
Pick and Place
Whiteboard Wiping
Table Cleaning
Instruction
Pick up the fruit and place it into the basket
Wipe all writings off the whiteboard
Clean the table by putting all objects into the basket
Appendix
Table 6: Real-world tasks and the natural-language instructions provided to the model.
Figure 6: Chunk-wise causal attention ablation on LIBERO. Removing the causal mask while keeping the same streaming buffer drops asynchronous success rate by roughly 10 points on average, isolating the chunk-wise causal mask as the active ingredient in FlashVLA’s asynchronous continuity.
Figure 7: Chunk size C ablation on LIBERO with buffer length N=5 fixed. (a) Success rate. (b) Per-step latency.
Figure 8: Buffer length N ablation on LIBERO with chunk size C=10 fixed. (a) Success rate. (b) Per-step latency.