❌

Normal view

Received — 8 April 2026 ⏭

Safety-Aligned 3D Object Detection: Single-Vehicle, Cooperative, and End-to-End Perspectives

✇cs.AI, q-bio.NC updates on arXiv.org
By: Brian Hsuan-Cheng Liao · Chih-Hong Cheng · Hasan Esen · Alois Knoll
7 April 2026 at 12:00
arXiv:2604.03325v1 Announce Type: cross Abstract: Perception plays a central role in connected and autonomous vehicles (CAVs), underpinning not only conventional modular driving stacks, but also cooperative perception systems and recent end-to-end driving models. While deep learning has greatly improved perception performance, its statistical nature makes perfect predictions difficult to attain. Meanwhile, standard training objectives and evaluation benchmarks treat all perception errors equally, even though only a subset is safety-critical. In this paper, we investigate safety-aligned evaluation and optimization for 3D object detection that explicitly characterize high-impact errors. Building on our previously proposed safety-oriented metric, NDS-USC, and safety-aware loss function, EC-IoU, we make three contributions. First, we present an expanded study of single-vehicle 3D object detection models across diverse neural network architectures and sensing modalities, showing that gains under standard metrics such as mAP and NDS may not translate to safety-oriented criteria represented by NDS-USC. With EC-IoU, we reaffirm the benefit of safety-aware fine-tuning for improving safety-critical detection performance. Second, we conduct an ego-centric, safety-oriented evaluation of AV-infrastructure cooperative object detection models, underscoring its superiority over vehicle-only models and demonstrating a safety impact analysis that illustrates the potential contribution of cooperative models to "Vision Zero." Third, we integrate EC-IoU into SparseDrive and show that safety-aware perception hardening can reduce collision rate by nearly 30% and improve system-level safety directly in an end-to-end perception-to-planning framework. Overall, our results indicate that safety-aligned perception evaluation and optimization offer a practical path toward enhancing CAV safety across single-vehicle, cooperative, and end-to-end autonomy settings.

RDEx-CMOP: Feasibility-Aware Indicator-Guided Differential Evolution for Fixed-Budget Constrained Multiobjective Optimization

✇cs.AI, q-bio.NC updates on arXiv.org
By: Sichen Tao · Yifei Yang · Ruihan Zhao · Kaiyu Wang · Sicheng Liu · Shangce Gao
7 April 2026 at 12:00
arXiv:2604.03708v1 Announce Type: cross Abstract: Constrained multiobjective optimisation requires fast feasibility attainment together with stable convergence and diversity preservation under strict evaluation budgets. This report documents RDEx-CMOP, the differential evolution variant used in the IEEE CEC 2025 numerical optimisation competition (C06 special session) constrained multiobjective track. RDEx-CMOP integrates an {\epsilon}-level feasibility schedule, a SPEA2-style indicator-driven fitness assignment, and a fitness-oriented current-to-pbest/1 mutation operator. We evaluate RDEx-CMOP on the official CEC 2025 CMOP benchmark using the median-target U-score framework and the released trace data. Experimental results show that RDEx-CMOP achieves the highest total score and the best overall average rank among all released comparison algorithms, with strong target-attainment behaviour and near-zero final violation on most problems.

XModBench: Benchmarking Cross-Modal Capabilities and Consistency in Omni-Language Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xingrui Wang · Jiang Liu · Chao Huang · Xiaodong Yu · Ze Wang · Ximeng Sun · Jialian Wu · Alan Yuille · Emad Barsoum · Zicheng Liu
7 April 2026 at 12:00
arXiv:2510.15148v2 Announce Type: replace-cross Abstract: Omni-modal large language models (OLLMs) aim to unify audio, vision, and text understanding within a single framework. While existing benchmarks primarily evaluate general cross-modal question-answering ability, it remains unclear whether OLLMs achieve modality-invariant reasoning or exhibit modality-specific biases. We introduce XModBench, a large-scale tri-modal benchmark explicitly designed to measure cross-modal consistency. XModBench comprises 60,828 multiple-choice questions spanning five task families and systematically covers all six modality compositions in question-answer pairs, enabling fine-grained diagnosis of an OLLM's modality-invariant reasoning, modality disparity, and directional imbalance. Experiments show that even the strongest model, Gemini 2.5 Pro, (i) struggles with spatial and temporal reasoning, achieving less than 60% accuracy, (ii) reveals persistent modality disparities, with performance dropping substantially when the same semantic content is conveyed through audio rather than text, and (iii) shows systematic directional imbalance, exhibiting lower consistency when vision serves as context compared to text. These findings indicate that current OLLMs remain far from truly modality-invariant reasoning and position XModBench as a fundamental diagnostic tool for evaluating and improving cross-modal competence. All data and evaluation tools will be available at https://xingruiwang.github.io/projects/XModBench/.

RASA: Routing-Aware Safety Alignment for Mixture-of-Experts Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jiacheng Liang · Yuhui Wang · Tanqiu Jiang · Ting Wang
7 April 2026 at 12:00
arXiv:2602.04448v2 Announce Type: replace-cross Abstract: Mixture-of-Experts (MoE) language models introduce unique challenges for safety alignment due to their sparse routing mechanisms, which can enable degenerate optimization behaviors under standard full-parameter fine-tuning. In our preliminary experiments, we observe that naively applying full-parameter safety fine-tuning to MoE models can reduce attack success rates through routing or expert dominance effects, rather than by directly repairing Safety-Critical Experts. To address this challenge, we propose RASA, a routing-aware expert-level alignment framework that explicitly repairs Safety-Critical Experts while preventing routing-based bypasses. RASA identifies experts disproportionately activated by successful jailbreaks, selectively fine-tunes only these experts under fixed routing, and subsequently enforces routing consistency with safety-aligned contexts. Across two representative MoE architectures and a diverse set of jailbreak attacks, RASA achieves near-perfect robustness, strong cross-attack generalization, and substantially reduced over-refusal, while preserving general capabilities on benchmarks such as MMLU, GSM8K, and TruthfulQA. Our results suggest that robust MoE safety alignment benefits from targeted expert repair rather than global parameter updates, offering a practical and architecture-preserving alternative to prior approaches.
Received — 4 April 2026 ⏭

Leptomeningeal metastatic cancer cells induce a permissive choroid plexus vasculature through extracellular-vesicle-derived 5-HIAA signaling

✇Nature Cancer
By: Yang Huang · Yifan Hou · Jiahui Yang · Wanqi Yang · Hanxin Liu · Haibao Peng · Xiaohui Li · Jiaxu Zhao · Ye Zhang · Rui Zeng · Youming Zeng · Cheng Li · Qiangqiang Zhang · Yudan Chi
3 April 2026 at 08:00

Nature Cancer, Published online: 03 April 2026; doi:10.1038/s43018-026-01145-y

Huang, Hou, Yang et al. demonstrate that leptomeningeal metastatic cells favor the formation of a premetastatic niche by remodeling the choroid plexus vasculature through the serotonin metabolite 5-hydroxyindoleacetic acid, which signals into endothelial cells through the aryl hydrocarbon receptor.
Received — 2 April 2026 ⏭

InfiniteVL: Synergizing Linear and Sparse Attention for Highly-Efficient, Unlimited-Input Vision-Language Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Hongyuan Tao · Bencheng Liao · Shaoyu Chen · Haoran Yin · Qian Zhang · Wenyu Liu · Xinggang Wang
1 April 2026 at 12:00
arXiv:2512.08829v2 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) are increasingly tasked with ultra-long multimodal understanding. While linear architectures offer constant computation and memory footprints, they often struggle with high-frequency visual perception compared to standard Transformers. To bridge this gap, we introduce \textbf{InfiniteVL}. We first develop a hybrid base model called \textbf{InfiniteVL-Base} that interleaves a small fraction of Full Attention layers with Gated DeltaNet. Empowered by a tailored distillation and fine-tuning strategy, InfiniteVL-Base matches the fundamental multimodal performance of equivalent Transformers while achieving a \textbf{1.7$\times$} decoding speedup. However, the quadratic complexity of the retained Full Attention inevitably becomes an efficiency bottleneck when scaling to ultra long context. To break this barrier, we propose a novel Long-Sequence Architectural Fine-Tuning strategy that seamlessly transforms the dense attention into vision-specific sparse mechanisms. This yields two specialized variants: \textbf{InfiniteVL-Offline} for offline retrieval and \textbf{InfiniteVL-Online} for online streaming. By eliminating the computation explosion of global attention without sacrificing high-frequency visual recall, InfiniteVL-Offline achieves Transformer-level length generalization with a \textbf{5x} prefill acceleration at 256K context. Concurrently, InfiniteVL-Online delivers robust streaming perception with a constant memory footprint and a real-time throughput of \textbf{25} FPS. Code and models are available at https://github.com/hustvl/InfiniteVL.

Proposed Role of Circadian Clock Genes in Pathogenesis of HCC: Molecular Subtyping and Characterization

✇Omics in Hepatocellular
By: Zhikui Lu · Yi Zhou · Jian Luo · Zhicheng Liu · Zhenyu Xiao
28 March 2026 at 18:00

Biomedicines. 2026 Mar 12;14(3):645. doi: 10.3390/biomedicines14030645.

ABSTRACT

Background: Hepatocellular carcinoma (HCC) stands as a prevalent global health issue with increasing incidence and mortality rates. Hepatocellular carcinoma (HCC) exhibits profound molecular and clinical heterogeneity, which limits the effectiveness of current therapeutic strategies. Circadian rhythm disruption has been implicated in metabolic reprogramming, proliferation, and immune modulation in cancer, but its role in shaping HCC heterogeneity remains poorly defined. Methods: Four public HCC transcriptomic cohorts (TCGA-LIHC, CHCC, LIRI, LICA) were integrated using RMA normalization and ComBat for batch correction. Consensus clustering based on 31 core circadian clock genes (CCGs) identified robust molecular subtypes. Multi-omics characterization-including genomic alterations, pathway activity (GSEA/GSVA), immune microenvironment profiling (CIBERSORT, EPIC, MCP-counter, xCell), and drug-sensitivity prediction (pRRophetic/oncoPredict)-was performed to delineate subtype-specific biological properties. A nine-gene CCG-based RiskScore model was constructed using LASSO Cox regression to internally validate subtype robustness and intra-subtype risk stratification. Results: Using consensus clustering of 31 core CCGs in TCGA-LIHC and three independent validation cohorts (CHCC, LIRI, LICA), we identified three reproducible subtypes-Cluster-1 (metabolic-quiescent), Cluster-2 (transition-intermediate), and Cluster-3 (proliferation-inflammatory)-which were recapitulated across cohorts and showed distinct overall survival (Cluster-3 worst; log-rank p values significant across datasets). Multi-omic characterization revealed that Cluster-3 exhibits the highest tumor mutational burden and CNV burden with enrichment of TP53/AXIN1/TERT alterations, strong activation of cell-cycle, E2F, and G2M programs, and an immune-hot yet immunosuppressed microenvironment enriched for TAMs, Tregs and MDSCs. By contrast, Cluster-1 shows relative genomic stability, dominant hepatic metabolic signatures (fatty-acid oxidation, bile-acid and xenobiotic metabolism) and an immune-cold phenotype. Single-cell mapping linked ALAS1 expression to malignant hepatocytes predominating in Cluster-1, whereas NONO and CSNK1D localized to stromal (CAFs/TECs) and both malignant/immune compartments respectively in Cluster-3, providing a cellular mechanism for subtype-specific metabolism, angiogenesis and immune modulation. Finally, a nine-gene CCG-based RiskScore validated prognostic stratification and drug-sensitivity predictions indicated subtype-specific therapeutic vulnerabilities (notably increased predicted TKI sensitivity in Cluster-3). Conclusion: In conclusion, this study proposes a robust circadian rhythm-based molecular classification of hepatocellular carcinoma, revealing three biologically and clinically distinct subtypes characterized by divergent genomic alterations, metabolic programs, immune microenvironment states, and prognostic patterns. By integrating bulk and single-cell transcriptomic data, we identify subtype-specific roles of key circadian regulators-including ALAS1, NONO, and CSNK1D-in shaping tumor metabolism, proliferation, stromal remodeling, and immune suppression. These findings highlight circadian dysregulation as a potential upstream factor associated with HCC heterogeneity and provide a conceptual framework for developing subtype-tailored mechanistic studies and circadian-informed therapeutic strategies.

PMID:41898292 | PMC:PMC13024568 | DOI:10.3390/biomedicines14030645

Received — 26 March 2026 ⏭

Think Before You Drive: World Model-Inspired Multimodal Grounding for Autonomous Vehicles

✇cs.AI, q-bio.NC updates on arXiv.org
By: Haicheng Liao · Huanming Shen · Bonan Wang · Yongkang Li · Yihong Tang · Chengyue Wang · Dingyi Zhuang · Kehua Chen · Hai Yang · Chengzhong Xu · Zhenning Li
25 March 2026 at 12:00
arXiv:2512.03454v3 Announce Type: replace-cross Abstract: Interpreting natural-language commands to localize target objects is critical for autonomous driving (AD). Existing visual grounding (VG) methods for autonomous vehicles (AVs) typically struggle with ambiguous, context-dependent instructions, as they lack reasoning over 3D spatial relations and anticipated scene evolution. Grounded in the principles of world models, we propose ThinkDeeper, a framework that reasons about future spatial states before making grounding decisions. At its core is a Spatial-Aware World Model (SA-WM) that learns to reason ahead by distilling the current scene into a command-aware latent state and rolling out a sequence of future latent states, providing forward-looking cues for disambiguation. Complementing this, a hypergraph-guided decoder then hierarchically fuses these states with the multimodal input, capturing higher-order spatial dependencies for robust localization. In addition, we present DrivePilot, a multi-source VG dataset in AD, featuring semantic annotations generated by a Retrieval-Augmented Generation (RAG) and Chain-of-Thought (CoT)-prompted LLM pipeline. Extensive evaluations on six benchmarks, ThinkDeeper ranks #1 on the Talk2Car leaderboard and surpasses state-of-the-art baselines on DrivePilot, MoCAD, and RefCOCO/+/g benchmarks. Notably, it shows strong robustness and efficiency in challenging scenes (long-text, multi-agent, ambiguity) and retains superior performance even when trained on 50% of the data.
Received — 14 March 2026 ⏭

SynPlanResearch-R1: Encouraging Tool Exploration for Deep Research with Synthetic Plans

✇cs.AI, q-bio.NC updates on arXiv.org
By: Hansi Zeng · Zoey Li · Yifan Gao · Chenwei Zhang · Xiaoman Pan · Tao Yang · Fengran Mo · Jiacheng Lin · Xian Li · Jingbo Shang
10 March 2026 at 12:00
arXiv:2603.07853v1 Announce Type: new Abstract: Research Agents enable models to gather information from the web using tools to answer user queries, requiring them to dynamically interleave internal reasoning with tool use. While such capabilities can in principle be learned via reinforcement learning with verifiable rewards (RLVR), we observe that agents often exhibit poor exploration behaviors, including premature termination and biased tool usage. As a result, RLVR alone yields limited improvements. We propose SynPlanResearch-R1, a framework that synthesizes tool-use trajectories that encourage deeper exploration to shape exploration during cold-start supervised fine-tuning, providing a strong initialization for subsequent RL. Across seven multi-hop and open-web benchmarks, \framework improves performance by up to 6.0% on Qwen3-8B and 5.8% on Qwen3-4B backbones respectively compared to SOTA baselines. Further analyses of tool-use patterns and training dynamics compared to baselines shed light on the factors underlying these gains. Our code is publicly available at https://github.com/HansiZeng/syn-plan-research.
Received — 11 March 2026 ⏭

CCR-Bench: A Comprehensive Benchmark for Evaluating LLMs on Complex Constraints, Control Flows, and Real-World Cases

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xiaona Xue · Yiqiao Huang · Jiacheng Li · Yuanhang Zheng · Huiqi Miao · Yunfei Ma · Rui Liu · Xinbao Sun · Minglu Liu · Fanyu Meng · Chao Deng · Junlan Feng
10 March 2026 at 12:00
arXiv:2603.07886v1 Announce Type: cross Abstract: Enhancing the ability of large language models (LLMs) to follow complex instructions is critical for their deployment in real-world applications. However, existing evaluation methods often oversimplify instruction complexity as a mere additive combination of atomic constraints, failing to adequately capture the high-dimensional complexity arising from the intricate interplay of content and format, logical workflow control, and real-world applications. This leads to a significant gap between current evaluation practices and practical demands. To bridge this gap, we introduce CCR-Bench, a novel benchmark designed to assess LLMs' adherence to complex instructions. CCR-Bench is characterized by: (1) deep entanglement of content and formatting requirements in task specifications; (2) instructions that involve intricate task decomposition, conditional reasoning, and procedural planning; and (3) evaluation samples derived entirely from real-world industrial scenarios. Extensive experiments on CCR-Bench demonstrate that even state-of-the-art models exhibit substantial performance deficiencies, clearly quantifying the gap between current LLM capabilities and the demands of realworld instruction understanding. We believe that CCR-Bench offers a more rigorous and realistic evaluation framework, advancing the development of LLMs toward the next generation of models capable of understanding and executing complex tasks in industrial applications.

Adaptation of Agentic AI: A Survey of Post-Training, Memory, and Skills

✇cs.AI, q-bio.NC updates on arXiv.org
By: Pengcheng Jiang · Jiacheng Lin · Zhiyi Shi · Zifeng Wang · Luxi He · Yichen Wu · Ming Zhong · Peiyang Song · Qizheng Zhang · Heng Wang · Xueqiang Xu · Hanwen Xu · Pengrui Han · Dylan Zhang · Jiashuo Sun · Chaoqi Yang · Kun Qian · Tian Wang · Changran Hu · Manling Li · Quanzheng Li · Hao Peng · Sheng Wang · Jingbo Shang · Chao Zhang · Jiaxuan You · Liyuan Liu · Pan Lu · Yu Zhang · Heng Ji · Yejin Choi · Dawn Song · Jimeng Sun · Jiawei Han
10 March 2026 at 12:00
arXiv:2512.16301v3 Announce Type: replace Abstract: Large language model (LLM) agents are moving beyond prompting alone. ChatGPT marked the rise of general-purpose LLM assistants, DeepSeek showed that on-policy reinforcement learning with verifiable rewards can improve reasoning and tool use, and OpenClaw highlights a newer direction in which agents accumulate persistent memory and reusable skills. Yet the research landscape remains fragmented across post-training, retrieval, memory, and skill systems. This survey studies these developments under a single notion of \emph{adaptation}: improving an agent, its tools, or their interaction after pretraining. We organize the field with a four-paradigm framework spanning agent adaptation and tool adaptation. On the agent side, A1 (tool-execution-signaled) and A2 (agent-output-signaled) improve the agent itself through supervised fine-tuning, preference optimization, and reinforcement learning with verifiable rewards. On the tool side, T1 (agent-agnostic) provides reusable pre-trained modules any agent can call, while T2 (agent-supervised) uses the agent's outputs to train memory systems, skill libraries, or lightweight subagents. Using this framework, we review post-training methods, adaptive memory architectures, and agent skills; compare their trade-offs in cost, flexibility, and generalization; and summarize evaluation practices across deep research, software development, computer use, and drug discovery. We conclude by outlining open problems in agent-tool co-adaptation, continual learning, safety, and efficient deployment.
Received — 5 March 2026 ⏭

Role-Aware Conditional Inference for Spatiotemporal Ecosystem Carbon Flux Prediction

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yiming Sun · Runlong Yu · Rongchao Dong · Shuo Chen · Licheng Liu · Youmi Oh · Qianlai Zhuang · Yiqun Xie · Xiaowei Jia
5 March 2026 at 13:00
arXiv:2603.03531v1 Announce Type: cross Abstract: Accurate prediction of terrestrial ecosystem carbon fluxes (e.g., CO$_2$, GPP, and CH$_4$) is essential for understanding the global carbon cycle and managing its impacts. However, prediction remains challenging due to strong spatiotemporal heterogeneity: ecosystem flux responses are constrained by slowly varying regime conditions, while short-term fluctuations are driven by high-frequency dynamic forcings. Most existing learning-based approaches treat environmental covariates as a homogeneous input space, implicitly assuming a global response function, which leads to brittle generalization across heterogeneous ecosystems. In this work, we propose Role-Aware Conditional Inference (RACI), a process-informed learning framework that formulates ecosystem flux prediction as a conditional inference problem. RACI employs hierarchical temporal encoding to disentangle slow regime conditioners from fast dynamic drivers, and incorporates role-aware spatial retrieval that supplies functionally similar and geographically local context for each role. By explicitly modeling these distinct functional roles, RACI enables a model to adapt its predictions across diverse environmental regimes without training separate local models or relying on fixed spatial structures. We evaluate RACI across multiple ecosystem types (wetlands and agricultural systems), carbon fluxes (CO$_2$, GPP, CH$_4$), and data sources, including both process-based simulations and observational measurements. Across all settings, RACI consistently outperforms competitive spatiotemporal baselines, demonstrating improved accuracy and spatial generalization under pronounced environmental heterogeneity.

Dual-Modality Multi-Stage Adversarial Safety Training: Robustifying Multimodal Web Agents Against Cross-Modal Attacks

✇cs.AI, q-bio.NC updates on arXiv.org
By: Haoyu Liu · Dingcheng Li · Lukas Rutishauser · Zeyu Zheng
5 March 2026 at 13:00
arXiv:2603.04364v1 Announce Type: cross Abstract: Multimodal web agents that process both screenshots and accessibility trees are increasingly deployed to interact with web interfaces, yet their dual-stream architecture opens an underexplored attack surface: an adversary who injects content into the webpage DOM simultaneously corrupts both observation channels with a consistent deceptive narrative. Our vulnerability analysis on MiniWob++ reveals that attacks including a visual component far outperform text-only injections, exposing critical gaps in text-centric VLM safety training. Motivated by this finding, we propose Dual-Modality Multi-Stage Adversarial Safety Training (DMAST), a framework that formalizes the agent-attacker interaction as a two-player zero-sum Markov game and co-trains both players through a three-stage pipeline: (1) imitation learning from a strong teacher model, (2) oracle-guided supervised fine-tuning that uses a novel zero-acknowledgment strategy to instill task-focused reasoning under adversarial noise, and (3) adversarial reinforcement learning via Group Relative Policy Optimization (GRPO) self-play. On out-of-distribution tasks, DMAST substantially mitigates adversarial risks while simultaneously doubling task completion efficiency. Our approach significantly outperforms established training-based and prompt-based defenses, demonstrating genuine co-evolutionary progress and robust generalization to complex, unseen environments.
Received — 25 February 2026 ⏭

DUET-VLM: Dual stage Unified Efficient Token reduction for VLM Training and Inference

✇cs.AI, q-bio.NC updates on arXiv.org
By: Aditya Kumar Singh · Hitesh Kandala · Pratik Prabhanjan Brahma · Zicheng Liu · Emad Barsoum
24 February 2026 at 13:00
arXiv:2602.18846v1 Announce Type: cross Abstract: Vision-language models (VLMs) have achieved remarkable multimodal understanding and reasoning capabilities, yet remain computationally expensive due to dense visual tokenization. Existing efficiency approaches either merge redundant visual tokens or drop them progressively in language backbone, often trading accuracy for speed. In this work, we propose DUET-VLM, a versatile plug-and-play dual compression framework that consists of (a) vision-only redundancy aware compression of vision encoder's output into information-preserving tokens, followed by (b) layer-wise, salient text-guided dropping of visual tokens within the language backbone to progressively prune less informative tokens. This coordinated token management enables aggressive compression while retaining critical semantics. On LLaVA-1.5-7B, our approach maintains over 99% of baseline accuracy with 67% fewer tokens, and still retains >97% even at 89% reduction. With this dual-stage compression during training, it achieves 99.7% accuracy at 67% and 97.6% at 89%, surpassing prior SoTA visual token reduction methods across multiple benchmarks. When integrated into Video-LLaVA-7B, it even surpasses the baseline -- achieving >100% accuracy with a substantial 53.1% token reduction and retaining 97.6% accuracy under an extreme 93.4% setting. These results highlight end-to-end training with DUET-VLM, enabling robust adaptation to reduced visual (image/video) input without sacrificing accuracy, producing compact yet semantically rich representations within the same computational budget. Our code is available at https://github.com/AMD-AGI/DUET-VLM.

ALOE: Action-Level Off-Policy Evaluation for Vision-Language-Action Model Post-Training

✇cs.AI, q-bio.NC updates on arXiv.org
By: Rushuai Yang · Hecheng Wang · Chiming Liu · Xiaohan Yan · Yunlong Wang · Xuan Du · Shuoyu Yue · Yongcheng Liu · Chuheng Zhang · Lizhe Qi · Yi Chen · Wei Shan · Maoqing Yao
24 February 2026 at 13:00
arXiv:2602.12691v2 Announce Type: replace-cross Abstract: We study how to improve large foundation vision-language-action (VLA) systems through online reinforcement learning (RL) in real-world settings. Central to this process is the value function, which provides learning signals to guide VLA learning from experience. In practice, the value function is estimated from trajectory fragments collected from different data sources, including historical policies and intermittent human interventions. Estimating the value function of current behavior quality from the mixture data is inherently an off-policy evaluation problem. However, prior work often adopts conservative on-policy estimation for stability, which avoids direct evaluation of the current high-capacity policy and limits learning effectiveness. In this paper, we propose ALOE, an action-level off-policy evaluation framework for VLA post-training. ALOE applies chunking-based temporal-difference bootstrapping to evaluate individual action sequences instead of predicting final task outcomes. This design improves effective credit assignment to critical action chunks under sparse rewards and supports stable policy improvement. We evaluate our method on three real-world manipulation tasks, including smartphone packing as a high-precision task, laundry folding as a long-horizon deformable-object task, and bimanual pick-and-place involving multi-object perception. Across all tasks, ALOE improves learning efficiency without compromising execution speed, showing that off-policy RL can be reintroduced in a reliable manner for real-world VLA post-training. Videos and additional materials are available at our project website.
There are no more articles
❌