DVAC uses denoising variance as an intrinsic signal to adaptively chunk actions in flow-based robot policies, improving success rates and cutting replans on LIBERO, RoboTwin, CALVIN, and real-world tasks.
hub Canonical reference
Mixture of horizons in action chunking
Canonical reference. 83% of citing Pith papers cite this work as background.
18
Pith papers citing it
Background
83% of classified citations
abstract
Vision-language-action (VLA) models have shown remarkable capabilities in robotic manipulation, but their performance is sensitive to the $\textbf{action chunk length}$ used during training, termed $\textbf{horizon}$. Our empirical study reveals an inherent trade-off: longer horizons provide stronger global foresight but degrade fine-grained accuracy, while shorter ones sharpen local control yet struggle on long-term tasks, implying fixed choice of single horizons being suboptimal. To mitigate the trade-off, we propose a $\textbf{mixture of horizons (MoH)}$ strategy. MoH rearranges the action chunk into several segments with different horizons, processes them in parallel with a shared action transformer, and fuses outputs with a light linear gate. It has three appealing benefits. 1) MoH exploits long-term foresight and short-term precision jointly within a single model, improving both performance and generalizability to complex tasks. 2) MoH is plug-and-play for full-attention action modules with minimal training or inference overhead. 3) MoH enables dynamic inference with adaptive horizons, which selects stable actions through cross-horizon consensus, achieving 2.5$\times$ higher throughput than baselines while preserving superior performance. Extensive experiments over flow-based policies $\pi_0$, $\pi_{0.5}$, and one-step regression policy $\pi_{\text{reg}}$ demonstrate that MoH yields consistent and significant gains on both simulations and real-world tasks. Notably, under mixed-task setting, $\pi_{0.5}$ with MoH reaches a new state-of-the-art with 99$\%$ average success rate on LIBERO after only $30k$ training iterations. Project page: https://timsty1.github.io/moh/
hub tools
citation-role summary
background 5
baseline 1
citation-polarity summary
years
2026 18representative citing papers
AutoSpeed learns annotation-free, stage-adaptive robot motion speeds by optimizing policies toward the minimum-cost DCT-retimed multi-speed demonstration target.
A modular benchmark of 100 dexterous manipulation tasks across 3 arms and 6 hands with 3,180 demonstrations reveals that current policies (Diffusion Policy, DP3, OpenVLA, π0.5) achieve only 34% mean success, exposing unsolved challenges in contact-rich and precise manipulation.
VLA-Corrector adds a detect-and-correct inference layer using a latent vision monitor and online gradient guidance to enable adaptive action horizons in chunked VLA policies.
PolicyTrim is an RL post-training framework that boosts VLA policy efficiency by 3x chunk utilization and 51.4% fewer steps, yielding up to 5.83x speedup.
Adding recurrent memory tokens to VLA models raises success rates on partially observable manipulation tasks from 0.42 to 0.84 on training and 0.07 to 0.23 on held-out tasks while preserving performance under full observability.
VLA models with inference-time steering mitigate action leakage in implicit human-robot collaboration, supporting longer horizons and yielding faster, more reliable assembly than shorter-horizon baselines in a 16-person study.
LIBERO and CALVIN fail multiple proposed diagnostics for shortcut solvability, statistical significance, overfitting, and data dependence, while a tiny 0.09B probe reaches near-SOTA on LIBERO.
PACE selects execution horizons online via low-speed valleys in predicted action chunks, boosting task success by 6.4 points in simulation and 19.7 points on real robots.
A3 adaptively sets a VLA model's execution horizon to the longest action prefix the model re-confirms under conditional and sequential re-decoding, matching or improving task success without manual tuning.
Learns state-conditioned commitment depth in a 7B vision-language policy that jointly predicts actions and replan intervals, outperforming fixed-depth baselines and larger models on Sliding Puzzle and Sokoban while providing a theoretical dominance result.
A verifier called Future Forward Dynamics Causal Attention enables adaptive action execution in World Action Models, reducing model inferences by 69% and improving success rates in robotic tasks.
AsyncShield restores VLA geometric intent from latency via kinematic pose mapping and uses PPO-Lagrangian to balance tracking with LiDAR safety constraints in a plug-and-play module.
SV-VLA uses infrequent heavy VLA planning of action chunks plus a lightweight closed-loop verifier to achieve both efficiency and robustness in dynamic robot control.
A two-stage framework pretrains an action module with temporal motion priors from unconditioned trajectories using flow-matching, then transfers it to VLA training via decoder reuse and distillation, yielding better performance on cross-embodiment tasks.
A frozen chunk-based robot policy plus a lightweight RL-trained horizon predictor raises success on high-precision and long-horizon manipulation by adapting open-loop length on the fly.
A single vision-language-action policy can execute robot manipulation at commanded speeds from 0.5x to 2x by training on merged/split demonstration actions conditioned on a speed scalar.
LingBot-VA combines video world modeling with policy learning via Mixture-of-Transformers, closed-loop rollouts, and asynchronous inference to improve robot manipulation in simulation and real settings.
citing papers explorer
-
Denoising Tells When to Replan: Denoising-Variance Adaptive Chunking for Flow-Based Robot Policies
DVAC uses denoising variance as an intrinsic signal to adaptively chunk actions in flow-based robot policies, improving success rates and cutting replans on LIBERO, RoboTwin, CALVIN, and real-world tasks.
-
AutoSpeed: Annotation-Free Stage-Adaptive Motion Speed Learning for Robot Manipulation
AutoSpeed learns annotation-free, stage-adaptive robot motion speeds by optimizing policies toward the minimum-cost DCT-retimed multi-speed demonstration target.
-
DexVerse: A Modular Benchmark for Multi-Task, Multi-Embodiment Dexterous Manipulation
A modular benchmark of 100 dexterous manipulation tasks across 3 arms and 6 hands with 3,180 demonstrations reveals that current policies (Diffusion Policy, DP3, OpenVLA, π0.5) achieve only 34% mean success, exposing unsolved challenges in contact-rich and precise manipulation.
-
VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon
VLA-Corrector adds a detect-and-correct inference layer using a latent vision monitor and online gradient guidance to enable adaptive action horizons in chunked VLA policies.
-
PolicyTrim: Boosting Intrinsic Policy Efficiency of Vision-Language-Action Models
PolicyTrim is an RL post-training framework that boosts VLA policy efficiency by 3x chunk utilization and 51.4% fewer steps, yielding up to 5.83x speedup.
-
$\mu$VLA: On Recurrent Memory for Partially Observable Manipulation in VLA Models
Adding recurrent memory tokens to VLA models raises success rates on partially observable manipulation tasks from 0.42 to 0.84 on training and 0.07 to 0.23 on held-out tasks while preserving performance under full observability.
-
Learning to Assist: Collaborative VLAs for Implicit Human-Robot Collaboration
VLA models with inference-time steering mitigate action leakage in implicit human-robot collaboration, supporting longer horizons and yielding faster, more reliable assembly than shorter-horizon baselines in a 16-person study.
-
What Are We Actually Benchmarking in Robot Manipulation?
LIBERO and CALVIN fail multiple proposed diagnostics for shortcut solvability, statistical significance, overfitting, and data dependence, while a tiny 0.09B probe reaches near-SOTA on LIBERO.
-
PACE: Phase-Aware Chunk Execution for Robot Policies with Action Chunking
PACE selects execution horizons online via low-speed valleys in predicted action chunks, boosting task success by 6.4 points in simulation and 19.7 points on real robots.
-
Dynamic Execution Commitment of Vision-Language-Action Models
A3 adaptively sets a VLA model's execution horizon to the longest action prefix the model re-confirms under conditional and sequential re-decoding, matching or improving task success without manual tuning.
-
When to Re-Commit: Temporal Abstraction Discovery for Long-Horizon Vision-Language Reasoning
Learns state-conditioned commitment depth in a 7B vision-language policy that jointly predicts actions and replan intervals, outperforming fixed-depth baselines and larger models on Sliding Puzzle and Sokoban while providing a theoretical dominance result.
-
When to Trust Imagination: Adaptive Action Execution for World Action Models
A verifier called Future Forward Dynamics Causal Attention enables adaptive action execution in World Action Models, reducing model inferences by 69% and improving success rates in robotic tasks.
-
AsyncShield: A Plug-and-Play Edge Adapter for Asynchronous Cloud-based VLA Navigation
AsyncShield restores VLA geometric intent from latency via kinematic pose mapping and uses PPO-Lagrangian to balance tracking with LiDAR safety constraints in a plug-and-play module.
-
Open-Loop Planning, Closed-Loop Verification: Speculative Verification for VLA
SV-VLA uses infrequent heavy VLA planning of action chunks plus a lightweight closed-loop verifier to achieve both efficiency and robustness in dynamic robot control.
-
Learning Action Priors for Cross-embodiment Robot Manipulation
A two-stage framework pretrains an action module with temporal motion priors from unconditioned trajectories using flow-matching, then transfers it to VLA training via decoder reuse and distillation, yielding better performance on cross-embodiment tasks.
-
Dynamic Execution Horizon Prediction for Chunk-based Robot Policies
A frozen chunk-based robot policy plus a lightweight RL-trained horizon predictor raises success on high-precision and long-horizon manipulation by adapting open-loop length on the fly.
-
TempoVLA: Learning Speed-Controllable Vision-Language-Action Policies
A single vision-language-action policy can execute robot manipulation at commanded speeds from 0.5x to 2x by training on merged/split demonstration actions conditioned on a speed scalar.
-
Causal World Modeling for Robot Control
LingBot-VA combines video world modeling with policy learning via Mixture-of-Transformers, closed-loop rollouts, and asynchronous inference to improve robot manipulation in simulation and real settings.