Text-Vision Synergistic Token Caching: A Training-Free Framework for Efficient Vision-Language-Action Inference

Qianer Li; Chengjie Zhang; Jingwen Chen; Zanjia Tong; Jiyuan Zhang; Hong Zhang

cs.CVSep 28, 2026

Text-Vision Synergistic Token Caching: A Training-Free Framework for Efficient Vision-Language-Action Inference

Authors: Qianer Li, Chengjie Zhang, Jingwen Chen, Zanjia Tong, Jiyuan Zhang, Hong Zhang

Organizations: College of Engineering, Southern University of Science and Technology (SUSTech), Shenzhen 518055, China · Master of Science in Data Science and Machine Learning, Department of Mathematics, Faculty of Science, National University of Singapore, Singapore

Abstract

Vision-Language-Action (VLA) models enable generalizable robotic control but remain computationally expensive. Token caching provides a training-free, plug-and-play acceleration alternative. However, existing VLA caching does not fully exploit a key inductive bias of VLA models: text-vision synergy, wherein textual semantics guide the precise visual grounding of task-relevant regions. In particular, existing designs insufficiently account for head-wise reliability in attention aggregation and layer-wise stability in cache reuse. To address this, we propose Text-Vision Synergistic Token Caching (TVCache), a training-free framework for efficient VLA inference. TVCache filters attention heads based on text-vision information focus to improve task-relevant and physically consistent visual grounding. Concurrently, we introduce a reuse-layer selection mechanism guided by text-vision entropy differences to avoid caching unstable representations and improve cache resource allocation. Extensive experiments across four representative VLA models, two simulation benchmarks, and real-world robotic tasks demonstrate the effectiveness and generality of TVCache. At matched token-retention ratios, TVCache consistently improves task success over existing VLA caching with comparable computational cost. On OpenVLA-OFT, it improves average success by up to 14.5 percentage points over VLA-Cache at 12.5% retention while reducing FLOPs by 2.45x relative to full-token inference.

Figures & tables

Fig. 1: Two bottlenecks of existing VLA token caching. (Left) Token-level spatial misalignment: indiscriminate attention-head aggregation introduces task-irrelevant responses and degrades visual grounding. (Right) Layer-level depth mismatch: reusing representations at unstable layers can propagate noisy cross-modal features, whereas delaying reuse after representations stabilize incurs redundant computation.

Fig. 2: Principle of KV cache reuse in VLA inference. At timestep $T$ , visual inputs are decoupled into static (blue) and dynamic (yellow) tokens. By reusing historical caches for invariant instructions and static tokens, the LLM allocates computation exclusively to new dynamic tokens, significantly accelerating autoregressive decoding.

Fig. 3: Overview of the proposed framework.

OpenVLA-OFT
Method	Spatial (%)	Object (%)	Goal (%)	Long (%)	Avg Success Rate (%) $\uparrow$	FLOPs (T) $\downarrow$	Control Freq. (Hz) $\uparrow$	Latency (ms) $\downarrow$
100% Tokens
Vanilla	99.00	98.00	97.20	94.40	97.15 (100.0%)	4.0139 (1.00 $\times$ )	10.43 (1.00 $\times$ )	95.87 (1.00 $\times$ )
Retain 50% Tokens
FastV	95.40	98.60	96.80	93.40	96.05 (98.9%)	2.4134 ( 1.66 $\times$ )	11.84 ( 1.13 $\times$ )	84.48 ( 1.13 $\times$ )
SparseVLM	96.20	88.40	96.20	75.80	89.15 (91.8%)	2.6809 ( 1.50 $\times$ )	11.71 ( 1.12 $\times$ )	85.41 ( 1.12 $\times$ )

TABLE I: Comparison of latency and performance between OpenVLA-OFT and BitVLA evaluated on the LIBERO Benchmark. The task success rate retention ratios (%) and speedup multipliers (

\times

) are calculated relative to the Vanilla baseline of each respective model at 100% tokens.

$\pi_{0.5}$
Method	Spatial (%)	Object (%)	Goal (%)	Long (%)	Avg Success Rate (%) $\uparrow$	FLOPs (T) $\downarrow$	Control Freq. (Hz) $\uparrow$	Latency (ms) $\downarrow$
100% Tokens
Vanilla	98.00	97.20	97.00	93.00	96.30 (100.0%)	2.1153 (1.00 $\times$ )	8.14 (1.00 $\times$ )	122.85 (1.00 $\times$ )
Retain 50% Tokens
FastV	57.00	67.20	60.80	59.40	61.10 (63.4%)	1.6061 ( 1.32 $\times$ )	8.90 ( 1.09 $\times$ )	112.34 ( 1.09 $\times$ )
SparseVLM	78.20	73.00	86.40	43.60	70.30 (73.0%)	1.5429 ( 1.37 $\times$ )	8.82 ( 1.08 $\times$ )	113.38 ( 1.08 $\times$ )

TABLE II: Comparison of latency and performance between

\pi_{0.5}

and VLA-Adapter evaluated on the LIBERO Benchmark.

VLA-Adapter
Method	Task 1 (%)	Task 2 (%)	Task 3 (%)	Task 4 (%)	Task 5 (%)	Avg Success Rate (%) $\uparrow$	FLOPs (T) $\downarrow$	Control Freq. (Hz) $\uparrow$	Latency (ms) $\downarrow$
100% Tokens
Vanilla	98.50	94.60	89.20	84.90	78.90	89.22 (100.0%)	0.2597 (1.00 $\times$ )	21.45 (1.00 $\times$ )	46.62 (1.00 $\times$ )
Retain 50% Tokens
VLA-Cache	95.60	91.00	84.30	76.80	69.40	83.42 (93.5%)	0.1831 ( 1.42 $\times$ )	22.72 ( 1.06 $\times$ )	44.01 ( 1.06 $\times$ )
TVCache (Ours)	97.10	92.50	87.90	81.10	73.10	86.34 (96.8%)	0.1751 ( 1.48 $\times$ )	22.73 ( 1.06 $\times$ )	43.99 ( 1.06 $\times$ )

TABLE III: Comparison of performance for VLA-Adapter evaluated on the CALVIN Benchmark.

Task Name	Description
SetKettle	Pick up the kettle and place it on the plate.
PackPear	Pick up the pear and place it inside the basket.
HangCup	Hang the cup on the wooden mug tree.
PotDuck	Put the duck into the pot and cover it with the lid.

TABLE IV: Descriptions of the evaluation tasks.

TVCache Ret.	Patch sim.	Head filtering	Token selection	Reuse-layer selection	Bookkeeping	Total (ms)
50%	0.742	0.746	0.322	1.012	0.066	2.888
25%	0.651	0.747	0.316	1.052	0.073	2.839
12.5%	0.782	0.708	0.304	0.968	0.075	2.837

TABLE V: Runtime overhead breakdown of TVCache under different token retention ratios.

Fig. 4: Real-world robot evaluation across four manipulation tasks. The figure shows the physical task scenarios and the corresponding success rates of TVCache and VLA-Cache under different token retention ratios.

OpenVLA-OFT
Method	Spatial	Object	Goal	Long	Avg. $\uparrow$
100% Tokens
Vanilla	99.00	98.00	97.20	94.40	97.15
Retain 50% Tokens
VLA-Cache	98.20	98.20	97.40	93.40	96.80
+ Head	98.60 ( $\uparrow 0.40$ )	98.40 ( $\uparrow 0.20$ )	97.80 ( $\uparrow 0.40$ )	93.80 ( $\uparrow 0.40$ )	97.15 ( $\uparrow 0.35$ )

TABLE VI: Ablation study on OpenVLA-OFT and BitVLA.

Fig. 5: Qualitative comparison of cross-modal attention. TVCache filters degraded heads, reducing background noise and spatial misalignment for more task-relevant visual grounding.

Explore similar work

Neural Introspection Gating for Adaptive KV-Cache Reuse in Vision-Language-Action Models
Aug 11, 2026Zhijie Wu, Kento Kawaharazuka, Kei OkadaDiffusion-Based Vision-Language-Actions Cache
Graduate School of Information Science and Technology, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan
Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement
Jul 7, 2026Ryuji Oi, Hikari Otsuka, Kosuke Matsushima +4Flow-Matching Vision-Language-Action Robotic Manipulation
Institute of Science Tokyo
VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models
Oct 6, 2026Owen Du, Yang Yue, Jie Zhang +3Diffusion-Based Vision-Language-Actions
ETH Zurich · Tsinghua University

Aug 11, 2026, cs.ROJ/K move · Enter open · S save

Neural Introspection Gating for Adaptive KV-Cache Reuse in Vision-Language-Action Models

Zhijie Wu, Kento Kawaharazuka, Kei Okada

Graduate School of Information Science and Technology, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan

Diffusion-Based Vision-Language-Actions Cache Introspection Adapters Reuse Autoregressive Transformers

Open paper AskPDF arXiv

Vision-Language-Action(VLA) models map camera images and language instructions directly to motor commands through a single autoregressive transformer. In real-time control, they still spend substantial compute recomputing key-value(KV) representations for visual tokens that barely change across neighboring frames. Recent work such as VLA-Cache reduces that cost by reusing KV states for visually static patches, but its policy relies only on observation-space heuristics and does not account for the model's own uncertainty. We propose Gated VLA-Cache, a lightweight, training-free extension that augments visual-similarity caching with neural introspection. The method monitors the logit margin between the top two predicted action tokens, a zero-cost confidence signal available during decoding. When the margin drops below a threshold, the cache is invalidated and a full recompute is triggered. Evaluated on four LIBERO benchmark suites with both OpenVLA and OpenVLA-OFT, Gated VLA-Cache improves reliability when blind caching hurts. On LIBERO-Goal and LIBERO-Long, it recovers over 100% of the lost accuracy while retaining 80% of the compute savings.

Abstract

Explore similar work

Neural Introspection Gating for Adaptive KV-Cache Reuse in Vision-Language-Action Models

Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement

VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models