Organizations: School of Computer Science and Engineering, Southeast University, Nanjing, China · Nanyang Technological University, Singapore · School of Cyber Science and Engineering, Southeast University, Nanjing, China · Purple Mountain Laboratories, Nanjing, China
Vision-Language-Action (VLA) models, as large foundation models for embodied control, have shown strong performance in manipulation tasks. However, their performance comes at high inference cost. To improve efficiency, recent methods adopt action chunking, which predicts a sequence of future actions for open-loop execution. Although effective for reducing computation, open-loop execution is sensitive to environmental changes and prone to error accumulation due to the lack of close-loop feedback. To address this limitation, we propose Speculative Verification for VLA Control (SV-VLA), a framework that combines efficient open-loop long-horizon planning with lightweight closed-loop online verification. Specifically, SV-VLA uses a heavy VLA as a low-frequency macro-planner to generate an action chunk together with a planning context, while a lightweight verifier continuously monitors execution based on the latest observations. Conditioned on both the current observation and the planning context, the verifier compares the planned action against a closed-loop reference action and triggers replanning only when necessary. Experiments demonstrate that SV-VLA combines the efficiency of chunked prediction with the robustness of closed-loop control, enabling efficient and reliable VLA-based control in dynamic environments. Code is available: https://github.com/edsad122/SV-VLA.
Figures & tables
Figure 1. Comparison of action chunking, speculative decoding, and our proposed Speculative Verification VLA (SV-VLA). (a) Action Chunking VLA predicts and executes an action chunk in an open-loop manner, so later actions may rely on stale observations. (b) Speculative Decoding VLA employs a draft model to generate candidate actions, and a heavy target model to verify them in parallel. However, the verified chunk is still executed based on stale observations. (c) Speculative Verification VLA (SV-VLA) performs chunk-level macro planning together with lightweight and frequent verification under online updated observations. During execution, it continuously verifies whether the current planned action remains valid under the latest observation and triggers replanning once a mismatch is detected. It combines the efficiency of chunk-level planning with the adaptability of closed-loop feedback, improving both responsiveness and robustness in dynamic environments. Comparison among Action-Chunking VLA, Speculative Decoding VLA, and our proposed Speculative Verification VLA.
Figure 2. Overview of SV-VLA. At each planning boundary T0 , a frozen heavy VLA takes the current observation I0 , language instruction L , and proprioceptive state s0 as input, and outputs a macro action chunk Amacro together with a planning context feature F0 . During execution, a lightweight verifier runs at control frequency. Given the latest observation I1 and the planning context feature F0 , it predicts a reference action a1′ , which is compared with the current planned action a1 from the macro chunk. If the discrepancy is below a threshold, the planned action a1 is accepted and executed; otherwise, it is rejected, the remaining chunk is discarded, and the heavy VLA replans from the current state. Overview of SV-VLA.
Method
Chunk Size
Success (%) ↑
Speed ↑
LIBERO-Goal
BASE ( K=8 ( Kim et al., 2025 ) )
8
96.8
1.00
BASE ( K=64 )
64
93.2
2.78
Speculative Decoding ( Wang et al., 2025 )
4
74.4
1.42
Speculative Verification (Ours)
64
94.4
1.56
LIBERO-Object
Table 1. Main results on the LIBERO benchmark. We compare different methods on three task suites. We report success rate and inference speed-up ratios relative to the open-loop baseline with chunk size K=8 . Bold indicates the best result, and underline indicates the second-best result. Overall, SV-VLA achieves a better balance between efficiency and robustness.
Metric
BASE ( K=8 )
BASE ( K=64 )
SV-VLA ( K=64 )
Macro-Planner calls / episode
14.5
4.2
6.7
Verifier calls/ episode
0
0
13.3
Average time / episode
15.9 s
5.7 s
8.8 s
Success rate
96.00%
79.47%
90.90%
Table 2. Inference time breakdown. We compare BASE ( K=8 and K=64 ) and SV-VLA in terms of average macro-planner/verifier calls per episode, average inference time per episode, together with success rate. Bold indicates the best result, and underline indicates the second-best result. Compared with BASE ( K=64 ), SV-VLA improves success through online verification and replanning.
Variant
Success (%) ↑
Speed ↑
Steps
Baselines
BASE ( K=8 )
93.0
1.00
8
BASE ( K=64 )
68.0
3.22
64
SV-VLA Variants
w/o context feature F0
73.7
1.46
24.9
w/o current observation It
63.7
1.19
13.1
Table 3. Ablation study of SV-VLA. We report success rate (Success), inference speed-up ratios (Speed), and the average number of executed action steps before each replanning event (Steps) on Libero-Spatial. SV-VLA with K=64 uses the default replanning threshold τ=0.2 . Bold indicates the best result, and underline indicates the second-best result.
Figure 3. Qualitative comparison on the task: “pick up the black bowl between the plate and the ramekin and place it on the plate.” SV-VLA detects during execution that the bowl has not been successfully grasped, interrupts the current macro action chunk, and replans from the latest observation, which enables successful task completion. In contrast, the open-loop baseline with K=64 executes the entire action chunk without correction and fails to recover from the grasping error, while the K=8 baseline remains robust through frequent replanning at a much higher inference cost.
Vision-language-action (VLA) policies commonly execute long-horizon mobile manipulation through open-loop action chunks, issuing multiple actions without receiving new high-level visual input. A committed chunk therefore implies how observations should evolve, but accidental deviations can violate this expectation while the remaining actions continue to propagate the error: commit-time policy confidence cannot react to a deviation that occurs after dispatch, and observation-only anomaly scores lack an action-conditioned reference for separating expected effects from unexplained changes. We propose CheckVLA, which verifies execution with a separately trained, frozen action-conditioned world model. A conformally calibrated risk threshold bounds the episode-level probability of an unnecessary first intervention and determines when to intervene, its exceedance controls how strongly the rewritten suffix retains the superseded chunk, latency-aware hard prefixing restricts replacement to actions that remain deployable, and an event-driven keyframe bank preserves evidence of prior progress across repairs. On RoboCasa365, under a common training recipe and a matched invocation budget, CheckVLA attains a 36.1% average success rate against 27.6% for periodic replanning (+8.5 points). At a matched 5% episode-level false-alarm target, action conditioning raises timely recall to 77.9%, against 48.6% for an observation-only control and 37.9% for an action-shuffled control. These simulation results support action-conditioned verification as a way to restore feedback during chunked execution while keeping the repair consistent with inference latency.
Yushan Liu, Peibo Sun, Xintao Chao +8
Tsinghua University · Shanghai Jiao Tong University · Peking University +2
Vision-Language-Action (VLA) foundation models have recently achieved strong progress in embodied intelligence. To reduce policy-call frequency while preserving temporal coherence, most generative policies adopt an action chunk mechanism, executing multiple future actions in an open-loop manner under a fixed action horizon. However, this "predict-then-blindly-execute" paradigm sacrifices closed-loop reactivity: in contact-rich physical interactions, even small local perturbations can rapidly amplify within the open-loop blind spot, leading to compounding errors and ultimately task failure. To address this limitation, we propose VLA-Corrector, a lightweight corrective inference framework for action-chunked VLA policies. Without modifying the backbone policy weights, VLA-Corrector introduces a lightweight Latent-space Vision Monitor (LVM) that continuously compares predicted and actual visual feature evolution, enabling online detection of visual dynamics deviations. Once persistent deviation is detected, the system triggers a truncation event, discards the remaining stale actions, and invokes corrective replanning via Online Gradient Guidance (OGG). The detect-and-correct mechanism of VLA-Corrector naturally induces an event-triggered adaptive action horizon: it preserves long-horizon execution when the current chunk remains reliable, and invokes short-horizon corrective replanning when execution begins to drift. In doing so, VLA-Corrector mitigates the trade-off imposed by static horizons between execution robustness and policy-call frequency. It can be integrated into different VLA models without further retraining the VLA backbone, interrupting compounding errors while preserving much of the efficiency benefit of action chunking and substantially improving robustness in long-horizon, contact-rich robotic manipulation tasks.
Current Vision-Language-Action (VLA) models typically treat the deepest representation of a vision-language backbone as universally optimal for action prediction. However, robotic manipulation is composed of many frequent closed-loop spatial adjustments, for which excessive abstraction may waste computation and weaken low-level geometric cues essential for precise control. Existing early-exit strategies attempt to reduce computation by stopping at predefined layers or applying heuristic rules such as action consistency, but they do not directly answer when a representation is actually sufficient for action. In this paper, we present LoopVLA, a recurrent VLA architecture that jointly learns representation refinement, action prediction, and sufficiency estimation. LoopVLA iteratively applies a shared Transformer block to refine multimodal tokens, and at each iteration produces both a candidate action and a sufficiency score that estimates whether further refinement is necessary. By sharing parameters across iterations, LoopVLA decouples refinement from absolute layer indices and grounds sufficiency estimation in the evolving representation itself. Since sufficiency has no direct supervision, we introduce a self-supervised distribution alignment objective, where intermediate confidence scores are trained to match the relative action quality across refinement steps, thereby linking sufficiency learning to policy optimization signals. Experiments on LIBERO, LIBERO-Plus, and VLA-Arena show that LoopVLA pushes the efficiency-performance frontier of VLA policies, reducing parameters by 45% and improving inference throughput by up to 1.7 times while matching or outperforming strong baselines in task success.
Boyang Shen, Kaixiang Yang, Hao Wang +4
1Huazhong University of Science and Technology · 2Wuhan United Imaging Surgical Co.,Ltd. (UIS)