Denoising
-
Breaking the Uniformity Trap: Scaling Video Diffusion Model via SplitMoE
2026-09-29 17:55:19Yu Xu, Yuxin Zhang, Xiao Yang, Haotian Yang, Yizhi Wang, Xinwei Huang, Minxuan Lin, Angtian Wang, Chongyang Ma, Fan TangAbstract
Mixture-of-Experts (MoE), popularized by large language models, is a promising paradigm for scaling visual generative models. However, conventional token-wise MoE routes tokens independently within a homogeneous expert pool and regularizes expert usage toward uniformity, making it poorly matched to video data that is spatiotemporally redundant and semantically long-tailed. We show that existing visual MoEs fall into a uniformity trap: semantically under-organized routing, compounded by uniform expert-usage regularization, scatters coherent patches across disparate experts, causing routing fragmentation and structural distortion. To address this, we propose SplitMoE, a split-role sparse architecture that breaks the shackles of uniformity. To accommodate the inherent semantic imbalance, we explicitly bifurcate the expert pool into semantic experts and generic experts, with semantic experts capturing high-level semantic abstraction and generic experts preserving residual visual information and flexible generative capacity. Leveraging prototype-guided routing and pull-push regularization, SplitMoE enables tokens to cluster naturally by semantic attributes rather than arbitrary balancing constraints. Extensive results show that under an equivalent activated-parameter budget, SplitMoE outperforms traditional load-balanced MoEs in convergence speed, routing coherence, and video generation quality across standard benchmarks. By revealing an emergent coarse-to-fine denoising logic, SplitMoE provides the community with a modality-aware scaling path, serving as a critical reference for building large-scale video world models.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38140
PDF
-
CLeaR: A Unified Framework for Resolving the Leakage-Degradation Dilemma in Style Transfer
2026-09-29 17:54:59Teng Zhou, Yunhao ChenAbstract
Style transfer aims to render target content in the style of a reference image, but existing methods often suffer from content leakage, where objects, layouts, or semantics from the style reference appear in the generated output. Although prior data-driven and training-free methods can reduce leakage, they often face a leakage-degradation dilemma: stronger content suppression may weaken style fidelity, while richer style preservation may reintroduce unwanted reference content. We identify this dilemma across the full style-transfer pipeline, including feature separation, feature-space grounding, and diffusion generation. To address these issues, we propose CLeaR, a training-free framework for content-leakage-resistant style transfer. CLeaR first uses Orthogonal Subspace Projection to define content-reduced style targets in each vision foundation model (VFM) feature space. It then performs Ensemble Inversion, which optimizes a shared pixel-space style anchor satisfying style constraints across multiple VFMs. Finally, Energy-Guided Calibration maintains style alignment during diffusion sampling by steering the denoising trajectory toward the ensemble-defined style manifold. We further provide a theoretical analysis showing that the style-anchor estimation error decreases with the number of VFMs. Experiments on StyleBench demonstrate that CLeaR improves style alignment, reduces content leakage, and achieves better LLM-as-Judge evaluation compared with existing methods. The code is available at \href{this https URL}{this https URL}.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38136
PDF
-
Self-Aligned Forcing: Streaming Video Diffusion with Differentiable Noisy History
2026-09-29 17:49:31Weiqiang Wang, Zhuokun Chen, Yusheng Dai, Boying Li, Yi Zhang, Hossein Rahmani, Qiuhong Ke, Jianfei CaiAbstract
Autoregressive video diffusion enables interactive streaming generation, but suffers from error accumulation over long rollouts. Self-rollout training reduces exposure bias, yet finite rollouts leave long-range drift unresolved. We observe that the noise level of the history key-value (K/V) representations trades visual quality against motion, and that restoring gradients through the history aligns causal training far more closely with bidirectional training. Motivated by these observations, we introduce Self-Aligned Forcing (SAF), a training scheme that aligns the history of each block with the noise level of the block being denoised. Specifically, the history is the K/V produced by preceding blocks at the same denoising stage, so all blocks at a stage can be denoised in a single forward pass under a causal mask. This keeps the noisy history differentiable, allowing future losses to optimize how it is encoded. SAF therefore avoids a separate no-gradient rollout and per-block timestep-zero recaching, training up to 1.8x faster than prior methods with lower memory. At inference, SAF achieves the highest single-GPU throughput among existing methods and keeps one history bank per stage for a multi-GPU pipeline, reaching 49.1 FPS on 4 GPUs. Experiments show superior long-horizon generation with a better balance between visual quality and motion. Project page: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38114
PDF
-
Residual Denoising Enables Sample-Efficient Multi-Agent Coordination on Demand
2026-09-29 16:55:13Dayi Dong, Maulik Bhatt, Aayushi Shrivastava, Lasse Peters, Negar MehrAbstract
Pretrained robot policies offer strong manipulation skills but are typically limited to single-agent settings, where a robot acts in isolation. In this work, we study how to adapt pretrained single-agent diffusion policies to multi-agent settings using minimal collaborative data, co-optimizing for two key objectives: high coordination performance and single-agent skill retention. To this end, we introduce ALTER, an adaptation method for coordination on demand: the adapted policy coordinates with other robots when deployed in a team while remaining capable of acting independently when operating alone. Execution is decentralized: each robot acts only on its own visual observations, without explicit inter-agent communication. Our method trains a coordination head that predicts a residual denoiser to transform single-agent behavior into coordinated multi-agent behavior when necessary while also preserving single-agent capabilities. To preserve single-agent capabilities, we augment a small number of collaborative demonstrations with self-distilled data generated by the base policy during training of the residual denoiser. In simulation, ALTER achieves higher coordination success over our baselines while retaining much higher source-skill retention. In our hardware experiments, we find similar trends where ALTER better co-optimizes for coordination success and single-agent skill retention than the baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2609.32129
PDF
-
SoL-Refiner: Speed-of-Light One-Step Refinement for High-Resolution Video
2026-09-29 16:37:29Haozhe Liu, Tian Ye, Shuchen Xue, Yitong Li, Junsong Chen, Haopeng Li, Jincheng Yu, Duomin Wang, Ruihua Zhang, Lei Zhu, Song Han, Enze XieAbstract
High-resolution video generation is expensive, as its cost grows rapidly with the number of spatiotemporal tokens. A practical alternative first generates a lower-resolution video and then applies a refiner, but conventional multi-step refinement introduces a second sampling bottleneck. We present SoL-Refiner, a one-step video refiner that transforms low-resolution model outputs into 4K videos with a single denoising step. Our three-stage recipe combines high-resolution continual training, reinforcement learning (RL) post-training, and a final one-step distillation. We introduce Refiner-Bench, a video refinement benchmark constructed from the outputs of different video generators, and use a shared-input protocol to compare refiners at approximately 2K output resolution. At 2K, the one-step SoL-Refiner outperforms all external refiners on the VBench and UniPercept averages, while at $3840\!\times\!2176$ it improves both metrics over the three-step LTX-2.3 Refiner. With the complete acceleration stack, SoL-Refiner achieves an $8.91\times$ speedup in refinement latency over the same baseline in our 2K latency setting.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37969
PDF
-
Time-Anchored Diffusion Language Models: Latent-Space Caching for Fast Generation
2026-09-29 16:16:26Joel Anto Paul, Litu Rout, Aditya Akella, Sanjay ShakkottaiAbstract
Recent work on anchored diffusion language models improves denoising by shaping an intermediate latent space with supervised important-token targets. In this work, we introduce time-based (self-supervised) anchoring, which learns and reuses latent anchors without requiring such targets. Our key observation is that anchors encode persistent properties of the clean sequence, such as its semantic intent, global structure, or intermediate plan. Although their hidden representations become stale as the token canvas evolves, their semantic content remains useful across nearby diffusion times. This is implemented through a two-stage architecture consisting of a relatively expensive anchor network that generates the latent cache state and a lightweight denoising network that intelligently combines the cached latent state with the current state at each reverse step using a fusion module. This gives anchoring a latent-space caching interpretation: the anchor network is evaluated periodically, while its cached representation is reused across multiple reverse steps. We instantiate this framework as TADM:Post-train, which time-anchorizes pretrained DLMs, and TADM:Pretraining, which learns time-based anchors during pretraining. Applied to DiffusionGemma-26B, TADM:Post-train improves throughput by approximately 49% to 79% on several math, code, and STEM benchmarks (GSM8K, AIME26, GPQA-Diamond, LiveCodeBench-v6, HumanEval, MMLU-Pro). TADM:Pretraining reduces Transformer-layer computation by up to 38% relative to a standard single-stage DLM, achieves up to 73% higher measured throughput than ADLM.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37924
PDF
-
2-Dimensional spectral gating for denoising bioacoustics recordings
2026-09-29 16:06:48Julien Boussard, M\'elisande Teng, Sulagna Saha, Mario Gallego-AbenzaAbstract
Isolating vocalizations from noise in bioacoustics recordings is a prerequisite to many ecological analyses, including species identification, animal communication understanding, and individual or population-level variability studies. However, when recordings are acquired in open environments, vocalizations, noise, or signal-to-noise ratio can vary widely across individuals, species, environment, and recording conditions, making it hard to develop robust and generalizable methods for ecological analyses. To account for these challenges, noise reduction techniques are used to remove noise before downstream analyses. Popular methods such as Noisereduce rely on spectral gating, which estimates a noise threshold for each frequency channel. We propose a further improvement to Noisereduce, leveraging the fact that most animal sounds have structure across multiple frequencies. We apply our method on bird and marine mammals recordings and show that our extension of Noisereduce leads to improved denoising in both above and under water acoustic recordings, without impacting speed of preprocessing.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37910
PDF
-
Urgent Actions Go First: Urgency-Aware Denoising for Real-Time VLA Control
2026-09-29 15:13:14Zibo Wang, Haochen Han, Pengzhen Ren, Mingtong Dai, Fangming LiuAbstract
Diffusion and flow-matching Vision-Language-Action (VLA) policies generate action chunks through iterative denoising, incurring substantial inference latency that severely limits real-time robotic control. Existing acceleration methods treat an action chunk as a monolithic computational unit, ignoring a crucial physical reality of receding-horizon control: actions are generated jointly but consumed sequentially, resulting in inherently heterogeneous execution urgencies. We exploit this asymmetry to introduce Urgency-Aware Denoising (UAD), a novel inference-time framework that allocates denoising computation according to when each action is physically needed. UAD releases time-critical urgent actions after fewer denoising steps while overlapping the continued background refinement of tail actions with physical execution. However, heterogeneous denoising introduces two key challenges: early-release errors in urgent actions and trajectory inconsistency in tail actions. UAD elegantly resolves both through two core mechanisms: Trajectory Reconciliation, which reconstructs unified internal state evolution to restore joint denoising coherence without additional model evaluations, and Ghost Action Correction, which leverages non-executed ghost continuations to dynamically compensate for early-release errors across remaining executable actions. Extensive evaluations across multiple VLA architectures, simulation benchmarks, and real-world manipulation tasks demonstrate that UAD achieves up to a 1.89x speedup in average action availability latency while maintaining comparable success rates to vanilla inference with optimal denoising budget, offering a more favorable success-latency trade-off than state-of-the-art VLA acceleration baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37772
PDF
-
E-MoE: Enhanced Mixture-of-Experts for Non-Factorized Diffusion Language Models
2026-09-29 13:18:50Arseny Ivanov, Alexander Kolesov, Alexander Korotin, Ivan Oseledets, Mikhail GoncharovAbstract
Masked diffusion models (MDMs) generate sequences by progressively unmasking several tokens per denoising step, but their reverse process is typically factorized over positions, limiting sample quality in the few-step regime where diffusion's speed advantage over autoregressive decoding matters most. A recent line of work introduces a continuous Gaussian latent, trained as a variational autoencoder, to capture correlations across positions, but such approaches are prone to posterior collapse, where the latent is silently ignored. We propose Enhanced Mixture-of-Experts (E-MoE), which builds the reverse process as a mixture of factorized distributions over a discrete shared latent given by the expert-routing decisions of a Mixture-of-Experts (MoE) backbone, without increasing active parameters over the factorized baseline. Across synthetic multi-modal benchmarks, binarized MNIST, and LM1B, E-MoE improves few-step generation over factorized baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37533
PDF
-
Remember What You Did: Action-History Memory with Dual-Expert Denoising for Long-Horizon Vision-Language-Action Policies
2026-09-29 11:43:19Yaxin Zhao, Dianye Huang, Chenwei Wang, Chenguang Yang, Zhongliang JiangAbstract
Vision-language-action (VLA) models have driven rapid progress in robotic manipulation, demonstrating strong fine-grained control and promising performance on long-horizon tasks. However, many existing VLAs lack explicit access to interaction history, making them vulnerable to perceptual aliasing: similar current observations and robot states at different task stages may induce action ambiguity and lower success rate. Existing methods incorporate temporal or progress cues through feature conditioning, action-prior modification, or sampling guidance. However, methods that jointly fine-tune memory modules and the base VLA incur additional policy-training costs, motivating the separation of trainable history-conditioned steering from frozen base-policy refinement. We propose ActMem-VLA, a dual-expert handover architecture that augments a frozen, fine-tuned VLA with a memory plugin comprising a Mamba-based memory module and a lightweight PreAction Expert (PAE). Specifically, Mamba encodes executed-action history into memory that conditions PAE alongside current context. With these inputs, PAE steers task progression during early, high-noise denoising, then passes the partially denoised action to the frozen Action Expert (AE) to refine action details during the remaining low-noise steps. The fine-tuned base VLA remains frozen throughout training, while only the Mamba module and PAE are jointly optimized. On LIBERO-Mem, ActMem-VLA achieves 80.8\% average success across all ten tasks, compared with 65.2\% for $\pi_{0.5}$ and 49.5\% for MemoryVLA, while introducing only 3.45\% additional parameters. Across four real-world tasks, it improves the average success rate over $\pi_{0.5}$ by 28.8\%.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37307
PDF
-
LDM-is-AE: Latent Diffusion Model is an Auto-Encoder for End-to-End Image Generation
2026-09-29 09:13:20Zhengqiang Zhang, Lingchen Sun, Rongyuan Wu, Qiaosi Yi, Xiangtao Kong, Chaodong Xiao, Lei ZhangAbstract
Latent Diffusion Models (LDMs) typically adopt a two-stage pipeline: an auto-encoder (AE) is first pre-trained to define a latent space, then a diffusion model is trained to perform denoising within it. Such a two-stage design introduces a representation mismatch, as the latent space is optimized for reconstruction rather than adapting the denoising dynamics. We reveal that the LDM itself is an AE, and consequently present LDM-is-AE, an end-to-end one-stage LDM training framework that eliminates the need for a separately trained tokenizer. Our key observation is that the LDM backbone actually performs a latent-to-feature-to-latent transformation at each denoising step, which can be interpreted as an internal decoding--encoding process. Leveraging this structure, we split the DiT backbone into two reciprocal components, DiT-E (i.e., DiT Encoding) and DiT-D (i.e., DiT Decoding), and impose image-space supervision on the intermediate features across all timesteps. Our model encourages the internal representation to align with the image domain throughout denoising, thereby establishing an explicit latent-to-image-to-latent path. At the zero-noise timestep, our model further performs an image-to-latent-to-image mapping, corresponding to an auto-encoding process. As a result, LDM-is-AE jointly learns latent representations and denoising dynamics in an end-to-end manner, yielding a diffusion-native latent space tailored to the generation process. Experiments demonstrate that LDM-is-AE exhibits highly competitive generation performance, achieving an FID of 1.80 and 1.90 on 256x256 and 512x512 class-conditional image generation, respectively.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37080
PDF
-
RAWD-TTS: Ratio-Free Reward Alignment for Discrete-Diffusion Voice Cloning
2026-09-29 08:42:33Maxim Maslov, Kirill Borodin, Vasilii Kudryavtsev, Nikita Vasiliev, Grach MkrtchianAbstract
Zero-shot text-to-speech synthesizes new utterances in a speaker's voice from a short reference recording. Voice cloning requires accurate content and preserved speaker identity, but supervised acoustic-token prediction does not directly optimize these waveform-level properties. Reward-based post-training addresses this mismatch, but in discrete diffusion, token choices and reveal positions jointly define the sampling trajectory, complicating alignment. We introduce RAWD-TTS (Ratio-free Advantage-Weighted Denoising), which scores decoded samples with recognition and speaker rewards and uses group-relative advantages to weight masked-token reconstruction of those samples, without reverse-trajectory likelihoods or target audio. On 500 Russian CV3-Eval voice-cloning prompts, joint alignment reduces word error rate from 3.18% to 2.42% at the reward-selected checkpoint (24.0% relative) and to 2.58% at the final checkpoint (19.0%), while WavLM speaker cosine rises from 0.733 to 0.748 and 0.755. Controlled experiments characterize recognition-identity trade-offs and the effects of corruption count, group composition, and weighting.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37028
PDF
-
World2Motion: Turning Video World Models into 3D Human Motion Generators
2026-09-29 08:30:04Tu Fangyuan, Xiangyue Zhang, Yiyi Cai, Yichen Peng, Kunhang Li, Bo Zheng, Zhixiang Wang, Kaipeng Zhang, Erwin Wu, Haoran Xie, Haiyang LiuAbstract
We present World2Motion, a framework that generates scene-aware 3D human motion and corresponding video from a single image and a text prompt. While existing 3D motion generators learn from motion datasets, their generalization is constrained by limited coverage of environments. In contrast, video world models such as Cosmos 3 offer broader environmental priors but are not designed for full-body motion generation; recovering motion from their generated videos requires costly two-stage inference. To address these, we turn Cosmos 3 into a single-stage 3D motion generator. This adaptation has two challenges: the scarcity of paired video--motion data and temporal instability in the generated motion. First, we construct a training dataset combining synthetic video--motion pairs with real videos paired with estimated 3D motion. Second, we propose a shift-decoupled noise schedule that assigns different noise levels to video and motion through shared denoising progress. This design accommodates the different denoising requirements of the two modalities, reducing motion jitter. Experiments on a multi-source interaction benchmark show that World2Motion has better motion--text alignment and scene interaction compared with the evaluated 3D motion generators. It also matches the interaction success rate of the two-stage baseline while achieving approximately 3.3$\times$ faster inference.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37004
PDF
-
Motion Concept Unlearning in Video Diffusion Models
2026-09-29 06:40:56Ping Liu, Chi ZhangAbstract
Text-to-video (T2V) diffusion models can generate realistic depictions of actions such as kicking, stabbing, and shooting, raising safety concerns that motivate targeted concept erasure. Although concept erasure has been extensively studied for static concepts in text-to-image and T2V models, erasing motion concepts remains largely unexplored. We present a systematic study of motion concept erasure in video Diffusion Transformers (DiTs). Through causal interventions, we show that text-conditioning attention carries concept-specific motion information and supports selective intervention, whereas perturbing temporal positional encoding suppresses both target and non-target dynamics. We further find that directly adapting ESD, a representative weight-level image erasure method, to a video DiT yields modest and uneven motion suppression: reducing its erasure training loss does not by itself remove the concept signal from the difference between the conditional and unconditional predictions, which classifier-free guidance (CFG) then scales at every denoising step. From these findings, we derive three requirements for motion concept erasure: concept specificity, spatial selectivity, and temporal naturalness. Each determines one component of MUTE (Motion concept Unlearning in Text-to-video gEneration): at each denoising step, MUTE extracts a concept direction through token neutralization, derives a spatial gate from the direction's intrinsic structure, and subtracts the resulting correction from the velocity output before CFG is applied. MUTE is training-free and requires no weight modification. Experiments on 20 motion concepts show that MUTE outperforms representative prompt-level, weight-level, and inference-time baselines on Wan2.1-T2V, and the same formulation transfers to CogVideoX, supporting its applicability across distinct T2V attention architectures.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36832
PDF
-
Learning via Self-Consistency for Diffusion-based Video Reasoning
2026-09-29 06:37:13Zhenghao Ni, Weimin Qiu, Meng TangAbstract
Video generation models have demonstrated emerging zero-shot capabilities for visual reasoning, perception, and other vision tasks. However, diffusion-based video generation is inherently stochastic, while many downstream vision tasks are deterministic. Motivated by the effectiveness of self-consistency in chain-of-thought reasoning for large language models, we investigate whether self-consistency can similarly improve diffusion-based video reasoning. We first introduce a training-free test-time scaling method that samples multiple video generations and aggregates their predictions through self-consistency. Specifically, we aggregate extracted paths, locations, or masks from multiple rollouts into a consensus prediction. To reduce the inference overhead of multi-rollout generation, we read out predictions early in the denoising trajectory, which preserves consensus quality while reducing denoising steps by more than half. We further propose Rejection Fine-Tuning (RFT) to distill consensus predictions into the video generation model. The resulting model internalizes the benefit of multi-sample consensus and requires only a single generation at inference time, while substantially outperforming the original model. Experiments on three tasks, including maze solving, visual search, and referring segmentation, show that both our self-consistency inference and consensus distillation dramatically improve video-based perception and reasoning, without requiring ground-truth videos or task-specific verification. For visual search, self-consistency raises task accuracy from 48.4% for a single generation to 99.0%. The distilled model retains much of the consensus benefit with a single rollout. For 4-by-4 maze solving, consensus-based training improves the single-generation strict success rate from 72.0% to 84.0% with the same inference latency.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36826
PDF
-
What Makes World Action Models Generalize? An Empirical Study of Test-Time Future Modeling
2026-09-29 03:39:32Renping Zhou, Zanlin Ni, Zihao Fan, Guohao Fu, Zeyu Liu, Hao Shi, Jie Zhang, Chi Bene Chen, Yang Yue, Xueyang Fu, Gao HuangAbstract
World action models (WAMs) predict the future alongside actions during training. Due to the heavy computation cost of video denoising, whether the future must still be generated during inference is disputed: Explicit WAMs denoise it into clean frames along with every action chunk, whereas Latent WAMs discard it entirely for acceleration. We find that latent WAMs, despite matching explicit ones on in-distribution tasks, fail to retain the generalization benefits that originally motivated WAMs. To demonstrate this, we evaluate generalization along three axes: environmental perturbation, data efficiency, and task generalization. Controlled comparisons with a matched backbone, training data, and budget reveal consistent degradation across all three axes when the action expert no longer conditions on future representations. Further analysis shows that the gap arises almost entirely from the first denoising step: the benefit comes from preparing the future, not generating it. We therefore propose Simple-WAM, which simplifies future modeling into a single forward pass of fully noised video tokens and adapts the training-time noise schedule to this inference behavior. Across simulation and real-world tasks, Simple-WAM achieves the best of both worlds, leading explicit WAMs in generalization performance with efficiency comparable to Latent WAMs. Project Page: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2609.34981
PDF
-
PulseQuant: Propagation-Guided Subspace Correction for 4-Bit Video Diffusion Transformers
2026-09-29 01:40:47Yutong Wang, Xingtong Ge, Enhuai Liu, Yunke Wang, Tianfan Xue, Xinyuan Chen, Chang XuAbstract
Quantization errors in video diffusion transformers can be amplified or attenuated by subsequent denoising updates, making local reconstruction error an incomplete predictor of final impact. We introduce PulseQuant, a 4-bit post-training quantization method that combines trajectory sensitivity with activation geometry to guide offline calibration. Isolated block--step interventions estimate propagation risk, which prioritizes sensitive trajectory states during row-radius selection. With these radii fixed, response-subspace correction uses neighboring-code edits to reduce residual components along dominant activation directions. Both stages preserve the original 4-bit weight representation. Controlled interventions show that short-horizon propagated error predicts final latent error more reliably than immediate block-output error, supporting calibration beyond local reconstruction objectives. Evaluations on Wan models, Self Forcing, and MiniMax-H3 demonstrate improvements in key consistency and dense-reference metrics while remaining competitive on other attributes across model scales and generation paradigms.
Abstract (translated)
URL
https://arxiv.org/abs/2609.33384
PDF
-
Staircase Policy: Streaming Inference for World-Action Models with Large Action Chunks
2026-09-29 01:32:04Guoheng Sun, Chen Chen, Jin Wang, Ang Li, Teresa LvAbstract
World-Action Models (WAMs) improve robotic manipulation by conditioning action generation on predicted future observations, but future prediction adds further inference overhead to already expensive iterative action generation. Action chunking can amortize this cost over multiple actions, yet performance degrades over long execution horizons because later actions remain conditioned on stale observations. We introduce STAIRCASE POLICY, a streaming inference and training framework that turns a flow-matching VLA into a JEPA-style WAM and partitions a large action chunk into sub-chunks at staggered denoising stages. Near-term actions are executed as soon as they become available, while later actions continue to be refined. At each sub-chunk boundary, the future latent is re-predicted from the latest observation and used to update all unexecuted actions, enabling long-horizon execution without repeated full policy inference. The resulting future-prediction error can further serve as a signal for adaptive chunking. S-WAM achieves 97.7% on LIBERO and 87.9% on LIBERO-Plus, and improves performance across multiple policy backbones and real-robot tasks. It reaches 292.7 executed actions per second, $3.62\times$ the throughput of conventional execution at comparable accuracy, while reducing time-to-first-action from 123.6 to 73.3 ms. With additional inference optimizations, throughput further increases to 642.9 actions per second.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36471
PDF
-
RA-CFGCache: From Branch-Level Criteria to Guided-Risk Control under Classifier-Free Guidance
2026-09-29 00:49:00Yiming Liu, Ben Wan, Tongxuan Liu, Ao Wang, Yuqi Xiong, Fan Zhang, Hui Chen, Guiguang DingAbstract
Diffusion models enable high-quality visual generation, but iterative denoising remains computationally expensive, especially under classifier-free guidance (CFG), which requires both conditional and unconditional evaluations. Training-free caching reduces this cost by reuse of previously computed features or predictions. However, existing branch-local reuse criteria do not explicitly account for how cache errors combine under CFG or how local perturbations affect the final output. We identify two misalignments in cache control: a branch-guided mismatch, where guided error depends on both the magnitudes and alignment of branch errors, and a local-final mismatch, where the downstream impact of a local error varies across timesteps. We propose RA-CFGCache, a Risk-Aligned Caching framework under CFG that incorporates both factors while keeping the sampling schedule and guidance rule fixed. CFG-aware Guided-Risk Composition combines existing branch-wise proxies using CFG coefficients and offline-calibrated cross-branch alignment. Propagation-Aware Rescaling further weights the resulting guided-risk estimate with a timestep-dependent propagation prior calibrated from isolated reuse perturbations. An online threshold controller then determines when to jointly refresh or reuse both branches. Experiments on FLUX.1-dev, Wan2.1-T2V-1.3B, and CogVideoX-2B demonstrate improved efficiency--fidelity trade-offs over evaluated training-free caching baselines. Moreover, RA-CFGCache is compatible with diverse base proxy families, including TeaCache-, DiCache-, and MagCache-style estimators, and consistently improves fidelity at nearly unchanged latency. Code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36433
PDF
-
Compress to Remember: Learning Compact Memory via On-Policy Distillation for Long Video Generation
2026-09-28 22:52:57Xiaoyu Wu, Weihang Guo, Yifei Wang, Xinze Feng, Lydia E. Kavraki, Zhiwei Steven WuAbstract
Standard video generators do not natively compact historical context into reusable memory tokens. As generation continues, the growing history makes it increasingly difficult to retain information from earlier frames due to long-context degradation. Key-frame-based approaches address this challenge by retaining selected past frames, but can discard information needed for future generation. Rather than relying on frame selection alone, we study whether a frozen video generator can supply the supervision needed to learn a compact representation of the history. We propose Prediction-Aligned Context Compaction (PACC), which uses a learned compressor to aggregate information across past frames into compact memory tokens. We train the compressor through on-policy distillation, using the same frozen generator both as a student when conditioned on compressed memory and as a teacher when conditioned on the full history. The student generates continuations, while the teacher provides targets for the same noisy inputs at each denoising step. Only the compressor is updated to align the student's predictions with these targets. We evaluate PACC on MBench, which jointly measures memory-event coverage and consistency. PACC outperforms the strongest baseline by 6.63 points on Causal-rCM and 3.19 points on Causal Forcing. Evaluation on VBench-Long using MovieGen prompts further shows that PACC produces minute-long videos with generation quality competitive with baselines. Together, these results show that learning to compact historical context can improve long-video memory without modifying the underlying generator.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36364
PDF
