Daily Papers Arch&EAI

Aerial vision-language navigation (VLN) requires an embodied agent to integrate visual evidence over time, plan future actions, and determine when it has reached a navigation goal under partial observability. Although recent VLA models offer a promising perception-to-action paradigm, adapting them to aerial navigation remains challenging due to limited historical context, short planning horizons, and unreliable implicit termination. To address these challenges, we propose DreamFly, a diffusion-based aerial VLN framework built on Dream-VLA. DreamFly introduces a causally aligned historical memory that augments the current visual representation using only observations preceding the current decision step, enabling temporal reasoning without future information leakage. We further formulate navigation as receding-horizon diffusion planning, where the policy predicts a $K$-step action chunk but executes only the first action before replanning. This plan-$K$, execute-one strategy uses future actions as auxiliary planning targets while preserving closed-loop visual feedback. Finally, LiteStop estimates the stop probability directly from action logits at the initial all-mask state, decoupling explicit termination from action generation. Experiments on the OpenFly benchmark demonstrate consistent improvements in seen and unseen environments. DreamFly achieves 32.04%/29.46% SR and 28.22%/23.54% SPL on the test-seen/test-unseen splits, respectively, outperforming all compared methods on both metrics while attaining the lowest navigation error. These results demonstrate the effectiveness of jointly modeling historical context, future action structure, and explicit termination for aerial VLN.

Aerial vision-language navigation (VLN) requires an embodied agent to integrate visual evidence over time, plan future actions, and determine when it has reached a navigation goal under partial observability.

Safe offline RL typically assumes access to dense per-step cost annotations, but in practice supervisors provide only trajectory-level stop-feedback: a binary signal at the first unsafe transition, with no per-step attribution. We frame this as a temporal credit assignment problem and propose the Redistribution-based Cost Inference (RCI) framework, which converts sparse stop-feedback into dense per-step costs via return decomposition, then trains a constrained offline policy on the augmented dataset. We show that return-equivalent redistribution preserves the feasible policy set and the optimal Lagrangian in a CMDP, establishing that the transformation is lossless in theory while yielding better-conditioned cost critic learning in practice. Experiments on highway driving and robotic manipulation demonstrate substantially lower violation rates than sparse and classifier-based baselines, with robustness to heterogeneous dataset compositions and label noise.

Safe offline RL typically assumes access to dense per-step cost annotations, but in practice supervisors provide only trajectory-level stop-feedback: a binary signal at the first unsafe transition, with no per-step attribution.

We present a novel visual servoing framework for controlling a robotic manipulator in configuration space using only natural visual features. To train our data-driven keypoint detector, we attach ArUco markers along the robot body, use their centers as keypoint labels, and apply image inpainting to remove the markers and reconstruct the occluded regions. This produces automatically labeled, markerless robot images without requiring accurate camera calibration or robot models. At runtime, a second inpainting model reconstructs robot regions that are partially occluded, enabling continuous keypoint detection. An Unscented Kalman Filter (UKF) further improves temporal consistency and robustness of the keypoint estimates. We demonstrate successful model-free, vision-based control using natural robot features under both full visibility and partial occlusion. To show broader applicability, we also extend the perception pipeline to two-module and three-module soft origami arms and qualitatively evaluate keypoint detection and temporal tracking on these platforms.

We present a novel visual servoing framework for controlling a robotic manipulator in configuration space using only natural visual features.

Robotic manipulation with dexterous hands is a cornerstone of Embodied AI, yet its progress is stifled by the high cost of collecting embodiment-aware teleoperation data. While abundant egocentric videos of human hands offer a scalable alternative, the profound discrepancies in appearance, articulation, and camera viewpoints between human and robotic data raise significant challenges for co-training. Though existing general image-editing models demonstrate strong capabilities, they lack necessary embodiment-specific priors to fully bridge this gap. In this work, we present HandEdit, a unified large-scale embodiment-aware image-editing dataset and benchmark specifically designed to transform human hands and arms into various dexterous robotic embodiments within egocentric frames. HandEdit comprises over 200M editing instances derived from five diverse source datasets, covering 26 distinct URDFs, including 13 hand-only and 13 hand-arm configurations. Alongside the dataset, we establish a unified benchmark protocol with two tracks: Hand-only and Hand-Arm, supporting URDF-conditioned evaluation. We conduct extensive evaluations of 11 representative image-editing baselines using a multi-dimensional metric suite, including generic similarity metrics, VLM-based judgment, and embodiment-aware metrics. HandEdit serves as a critical resource at the intersection of image editing and robotics: it advances embodiment-aware editing models while enabling scalable dexterous robotic learning from abundant human video data, paving the way for more generalizable Embodied AI.

Robotic manipulation with dexterous hands is a cornerstone of Embodied AI, yet its progress is stifled by the high cost of collecting embodiment-aware teleoperation data.

3D Visual Grounding (3DVG) localizes objects from natural-language descriptions in 3D scenes and is fundamental to embodied AI applications. Although foundation models enable open-vocabulary reasoning, they typically rely on pre-generated candidates, creating two sequential bottlenecks. The \emph{candidate bottleneck} occurs when dataset-specific 3D proposal models miss, fragment, or incorrectly group targets under distribution shifts, excluding them from VLM reasoning. The \emph{evidence bottleneck} stems from incomplete visual evidence: global renderings preserve spatial context but obscure object details, whereas candidate-centric views capture local appearance but lack global context. To address these bottlenecks, we propose UniGround, a zero-shot 3DVG framework that addresses both bottlenecks through Global Candidate Filtering and Contextual Precision Grounding. Global Candidate Filtering constructs topology-consistent, class-agnostic candidates from 3D topology and multi-view semantic cues, without dataset-trained 3D detectors, task-specific proposal supervision, or predefined box and category priors. Contextual Precision Grounding jointly reasons over global spatial context and candidate-centric visual evidence, followed by closed-loop consistency verification for reliable target identification. UniGround achieves 46.1\%/34.1\% Acc@0.25/0.5 on ScanRefer and 28.7\% Acc@0.25 on the evaluated ARKitScenes subset of EmbodiedScan. Further experiments demonstrate competitive grounding without dataset-specific 3D priors, cross-dataset generalization to unseen indoor scenes, and robustness to real-world reconstruction noise and practical domain shifts.

3D Visual Grounding (3DVG) localizes objects from natural-language descriptions in 3D scenes and is fundamental to embodied AI applications.

Humans naturally leverage diverse sensing modalities to interact with the physical world, while most Vision-Language-Action (VLA) models for robotics rely solely on RGB observations. This limits their ability to perceive physical properties that are difficult or impossible to infer from RGB cameras, such as temperature, sound, or radar response. We present MuseVLA, an adaptive multimodal sensing VLA model that integrates novel sensors as on-demand tools for robotic manipulation. Given a task instruction and visual context, MuseVLA first generates a sensor token and target description that select the sensing modality to invoke and what to attend to, analogous to a tool call with arguments. It then converts the selected sensor measurement into a grounded sensor image, a unified intermediate representation that encodes heterogeneous readings for multimodal fusion and action generation. This design decouples sensor-specific processing from the VLA backbone, enabling efficient integration of diverse modalities. To reduce the need for expensive multisensory robot datasets, we further introduce a data synthesis pipeline that augments existing RGB video datasets with grounded sensor images, enabling generalization to unseen sensor-guided tasks. We evaluate MuseVLA on a real-world robot across challenging dexterous hand manipulation tasks that require multimodal sensing inputs, including temperature-guided pick-and-place, audio-driven object search, and radar-assisted hidden object retrieval. MuseVLA achieves 80.6% success rate on average, outperforming RGB-only and multisensory VLA baselines significantly, and exhibits strong zero-shot capabilities on unseen tasks. Code, model and dataset are available at https://github.com/microsoft/MuseVLA.

Humans naturally leverage diverse sensing modalities to interact with the physical world, while most Vision-Language-Action (VLA) models for robotics rely solely on RGB observations.

Vision-and-Language Navigation (VLN) has progressively expanded from indoor to outdoor environments. However, existing outdoor VLN datasets still rely on fixed discrete topological graphs for construction. It fails to align with the rapidly changing real-world outdoor environments and impedes the sim-to-real transfer of VLN agents. To address this limitation, we propose DaViNCi (\textbf{D}yn\textbf{a}mic \textbf{Vi}sion-and-Language \textbf{N}avigation in \textbf{C}ont\textbf{i}nuous Environment), the first outdoor VLN dataset that simultaneously introduces both continuous and dynamic factors. The agent not only moves in the outdoor environment using continuous actions but is also required to handle unpredictable dynamic elements. The dataset encompasses six distinct maps with a total of 6,933 trajectories. Through comprehensive comparative experiments, we find that the success rate on DaViNCi decreased by more than 10\% in discrete environments compared to previous datasets. And there is an even greater decline in continuous settings, demonstrating the challenge of DaViNCi. Furthermore, we clarify the impact of action granularity and dynamic elements. These results demonstrate the practical value of DaViNCi in advancing outdoor VLN toward more realistic environments. The website is https://xzh0312.github.io/DaViNCi/.

Vision-and-Language Navigation (VLN) has progressively expanded from indoor to outdoor environments.

Test-time training (TTT) offers a lightweight way to adapt vision--language--action (VLA) policies from unlabeled deployment streams, but it remains difficult to use reliably in closed-loop manipulation. A shared adaptation space can mix incompatible task corrections, while an online update can alter subsequent actions before its consequences are known. We introduce a reliable TTT framework for VLA policies (VANE). VANE conditions prompt adaptation on the current vision--language context and learns from the future visual consequences of executed actions. Candidate updates are isolated from the live policy, evaluated on subsequent observations, and committed only when supported by future evidence, making adaptation selective and reversible. On SimplerEnv WidowX, VANE improves average success by $3.2$ percentage points over the corresponding TTT baseline. Results on Google Robot further show that deployment-time gains remain task- and embodiment-dependent. Together, these results demonstrate a constrained, evidence-based approach to adapting VLA policies during interaction.

Test-time training (TTT) offers a lightweight way to adapt vision--language--action (VLA) policies from unlabeled deployment streams, but it remains difficult to use reliably in closed-loop manipulation.

Training and validation of Embodied AI for social navigation critically depends on realistic simulation environments, yet many current approaches fail to find a balance between realism and simulability. We propose D3D-GEN, a novel world generation system that combines a domain agent with a retrieval-augmented generation (RAG) pipeline grounded in that domain. Our system enables users to rapidly generate domain-grounded, fully interactive 3D worlds by automating both the collection of domain knowledge and the synthesis of realistic floorplans and object placements, without dependence on any fixed 3D model database. Given a domain description prompt, the research agent collects publicly accessible domain-specific data and constructs a persistent domain database. Using this database, our RAG pipeline generates plausible floorplans and object placements by dynamically querying a user-provided semantic database, which can be easily extended or modified. The output is a fully interactive 3D world loadable by the popular simulators Isaac Sim and Gazebo. With our approach, we have built databases for several common domains (indoor residential, hospital, office) and generated dozens of distinct, plausible simulation environments for each domain. We present D3D-GEN with a local web frontend that facilitates rapid, interactive world generation for robot simulation.

Training and validation of Embodied AI for social navigation critically depends on realistic simulation environments, yet many current approaches fail to find a balance between realism and simulability.

On-device skin lesion analysis is constrained by the compute and energy cost of conventional CNN inference and by the need for lightweight calibration under clinical data shift. Neuromorphic processors provide event-driven sparse computation, but practical deployment is often limited by CNN-to-SNN conversion failures, including unsupported operators, quantization distortion, and accuracy degradation under class imbalance. We propose QANA, a quantization-aware CNN backbone within an end-to-end pipeline for conversion-stable neuromorphic execution. QANA improves conversion robustness by bounding intermediate activations, aligning normalization with low-bit quantization, and replacing conversion-fragile components with spike-compatible transformations. Efficient representation is achieved through Ghost-based feature generation, spatially-aware efficient channel attention, and squeeze-and-excitation modules whose operations can be quantized, folded, or lowered into integer graph operations. The quantized projection head produces SNN-ready logits and supports lightweight readout calibration without full retraining or data offloading. On HAM10000, QANA achieves 91.6% Top-1 accuracy and 91.0% macro F1, improving the strongest converted SNN baseline by 3.5 points in accuracy and 12.1 points in macro F1. On a clinical dataset, QANA achieves 90.8% Top-1 accuracy and 81.7% macro F1, improving the strongest baseline by 3.2 points in accuracy and 3.6 points in macro F1. On BrainChip Akida, QANA runs in 1.5 ms with 1.7 mJ per image, outperforming the strongest size-matched Akida baseline by 4.1 points in accuracy, 10.0 points in macro F1, 21.1% in latency, and 22.7% in energy.

On-device skin lesion analysis is constrained by the compute and energy cost of conventional CNN inference and by the need for lightweight calibration under clinical data shift.

At its core, robotic manipulation is a problem of vision-to-geometry mapping ($f(v) \rightarrow G$). Physical actions are fundamentally defined by geometric properties like 3D positions and spatial relationships. Consequently, we argue that the foundation for generalizable robotic control should be a vision-geometry backbone, rather than the widely adopted vision-language or video models. Conventional VLA and video-predictive models rely on backbones pretrained on large-scale 2D image-text or temporal pixel data. While effective, their representations are largely shaped by semantic concepts or 2D priors, which do not intrinsically align with the precise 3D geometric nature required for physical manipulation. Driven by this insight, we propose the Vision-Geometry-Action (VGA) model, which directly conditions action generation on pretrained 3D representations. Specifically, VGA replaces conventional language or video backbones with a pretrained 3D world model, establishing a seamless vision-to-geometry mapping that translates visual inputs directly into physical actions. To further enhance geometric consistency, we introduce Progressive Volumetric Modulation and jointly train action and 3D property prediction to preserve geometric representations. Extensive experiments validate the effectiveness of our approach. Across simulation benchmarks, VGA outperforms leading VLA, 3D-VLA, and WAM baselines, including $π_{0.5}$, OpenVLA-OFT, GeoVLA, and Motus. In real-world deployments, VGA surpasses $π_{0.5}$ under unseen viewpoints and accurately follows language instructions for target grasping. These results highlight that operating on native 3D representations, rather than relying primarily on language or video priors, offers a promising direction toward generalizable physical intelligence. Project page: https://hcplab-sysu.github.io/VisionGeometryActionModel.

At its core, robotic manipulation is a problem of vision-to-geometry mapping ($f(v) \rightarrow G$).

Recent studies have demonstrated the potential of reinforcement learning (RL) to improve the task performance of vision-language-action (VLA) models through interaction. However, current efforts remain fragmented, lacking a unified platform for fair comparison across architectures and algorithms, as well as an efficient system design for scalable training. Therefore, we present RLinf-VLA, a unified and efficient framework for scalable RL training of VLA models. RLinf-VLA standardizes the integration of diverse VLA architectures, RL algorithms, and heterogeneous simulators through a unified interface, enabling extensibility and reproducibility. To improve efficiency, the framework adopts a flexible resource allocation architecture for rendering, inference, and training in RL pipelines. In particular, RLinf-VLA introduces a hybrid fine-grained pipeline allocation strategy that achieves a 1.61$\times$-1.88$\times$ training speedup on ManiSkill. Using this framework, RL-trained models achieve strong performance across embodied benchmarks, including 98.11% success on 130 LIBERO tasks, 97.66% success on 25 ManiSkill tasks, and 84.63% average success across 6 RoboTwin tasks. In addition, RLinf-VLA distills a set of effective practices for RL-based VLA training. We envision RLinf-VLA as a foundational framework for efficient, unified, and reproducible research in embodied intelligence.

Recent studies have demonstrated the potential of reinforcement learning (RL) to improve the task performance of vision-language-action (VLA) models through interaction.

Vision-language-action (VLA) policies are expected to operate robustly across variations in the robot's initial configuration, yet aggregate task success can conceal pose-specific failures and inappropriate hand selection. This work investigates initial-pose dependence in VLA-based humanoid dual-arm manipulation. We characterize the initial-condition-dependent early hand preference as a policy-induced hand prior and quantify it using HandPriorScore, residual hand bias, and target responsiveness. Evaluations across multiple policies and 17 initial configurations reveal strong initial-pose--policy interactions: the same pose produces substantially different success rates across policies, while a single policy exhibits large performance variation across poses. Specific initial arm configurations can suppress or induce an asymmetric hand preference, with the resulting effect varying in direction and strength across policies. Wrist-camera observations also influence hand selection and task performance. Expanding initial-pose coverage in the training dataset substantially improves robustness, while targeted augmentation around a low-performing configuration increases its success rate. Comparisons across training configurations show that sufficient exposure to the target simulation task is beneficial, whereas the effect of real or auxiliary data depends on pose coverage, simulation ratio, and observation availability. These findings characterize a pose-conditioned hand prior, identify a localized initial arm configuration as a causal handle on hand-selection behavior, and demonstrate how data coverage and training composition affect initial-pose robustness.

Vision-language-action (VLA) policies are expected to operate robustly across variations in the robot's initial configuration, yet aggregate task success can conceal pose-specific failures and inappropriate hand selection.

Despite the promise of Vision-Language-Action (VLA) models as generalist robotic controllers, their robustness against perceptual noise and environmental variations in out-of-distribution (OOD) tasks remains fundamentally limited by the absence of long-term memory, causal failure attribution, and dynamic intervention capability. To address this, we propose RoboHarness, a memory-augmented policy harness that upgrades frozen VLA policies for robust in-context adaptation without parameter fine-tuning. Specifically, RoboHarness operates through an online pipeline of contrastive Dual-Memory Retrieval-Augmented Generation (RAG), an attribution-driven vision-language orchestrator implemented with a multimodal large language model, and extensible Model Context Protocol (MCP) interventions, while an offline Memory Consolidation module continuously distills the execution traces into reliable priors. Experimental evaluations across three backbone models ($π_0$, $π_{0.5}$, and SmolVLA) on LIBERO-PRO and our proposed LIBERO-RoboHarness benchmarks demonstrate that RoboHarness achieves an average absolute success rate gain of 56.6%. This includes a significant absolute improvement of 89.1% in long-horizon task chaining. The project page and source code are available at https://github.com/LZY-1021/RoboHarness.

Despite the promise of Vision-Language-Action (VLA) models as generalist robotic controllers, their robustness against perceptual noise and environmental variations in out-of-distribution (OOD) tasks remains fundamentally limited by the absence of long-term memory, causal failure attribution, and dynamic intervention capability.

The prevailing recipe for Vision-Language-Action (VLA) models couples a pretrained VLM with a separately trained flow-matching action expert. This makes the VLM a context encoder rather than a decision-maker. We introduce G0.5, a pretrained autoregressive VLA in which a single transformer decoder emits reasoning and action tokens under a single objective. Three components make this tractable at foundation-model scale: a learnable cross-embodiment action tokenizer that maps heterogeneous robot actions into a shared vocabulary; a native chain-of-thought stream interleaving task decomposition, object grounding, and action hints with action tokens; and a visual memory module that injects multi-second history through the vision encoder. Because reasoning and action share a single set of weights, the pretrained VLM's capabilities carry over to physical behavior: the model follows instructions closely, and prompts directly steer action granularity, task horizon, and out-of-distribution scene handling without further training. Pretrained on a large collection of robot datasets together with VQA samples, G0.5 surpasses state-of-the-art models across 7 independent regimes: real-world fine-tuning on R1lite and R1pro robots (76.7\% vs.\ 53.3\% for $π_{0.5}$ and 24.4\% for GR00T-N1.7), the 2025 BEHAVIOR Challenge on 50 long-horizon household mobile manipulation tasks using a generalist policy (31.4\% vs.\ 26.3\% for $π_{0.5}$ and 26.1\% for the challenge winner), DROID post-training followed by zero-shot transfer to an unseen environment and objects (82.5\%), a language-following Pick-and-Place benchmark, LIBERO (98.9\%), RoboTwin 2.0 (93.3\%), and SimplerEnv-Bridge (87.3\%).

The prevailing recipe for Vision-Language-Action (VLA) models couples a pretrained VLM with a separately trained flow-matching action expert.

Autonomous logistics sorting systems (ALSS) are an important industrial application of embodied AI, which requires joint planning over spatially disjoint camera views. We formulate this setting as Joint Multi-Scene Understanding (JMSU). With open-world visual understanding and task-planning capabilities, vision-language models (VLMs) are promising candidates for JMSU. However, directly applying existing VLMs to JMSU is non-trivial due to scarce cross-scene supervision and attention dispersion caused by long visual context in JMSU. To address these challenges, we propose HUGIN, a training framework with two complementary components. Endogenous Data Augmentation recombines verified atomic facts under operating constraints, while Global Context Ranking aligns the instruction representation more strongly with the complete visual context than with a partial visual context. To support ongoing research, we construct a high-quality industrial sorting dataset and benchmark named SortingBench from four layouts of autonomous logistics sorting systems. Across five open VLMs, HUGIN consistently outperforms matched baselines; for example, the accuracy on SortingBench of Qwen3-VL-8B increases from 63.6% to 78.8%. Additional experiments verify the effectiveness of each component and JMSU's spillover benefits in embodied tasks. Deployment tests involving more than 15,000 packages support the practical viability of VLM-based planning for autonomous logistics sorting.

Autonomous logistics sorting systems (ALSS) are an important industrial application of embodied AI, which requires joint planning over spatially disjoint camera views.

Vision-Language-Action (VLA) models can follow instructions and manipulate objects, but their performance often collapses out of distribution (OOD), when the scene, viewpoint, or object differs from training. Adapting to each new situation typically requires collecting more data and fine-tuning. We present StellaVLA, a framework that instead adapts at test time by conditioning on a single retrieved demonstration. The key idea is to move beyond imitating what an expert did and instead convey why: an automated offline pipeline converts each raw trajectory into a structured demonstration, e.g., a task plan, sub-goal descriptions, and verbalized 3D motion, at zero human-annotation cost. Provided as in-context guidance, this structured demonstration lets the policy reason about the task rather than mimic a pixel trajectory, which also makes it transferable across embodiments (real-robot, human-hand, or XR demonstrations). A parallel dual-training design internalizes this reasoning during training through a joint action-and-language objective, while inference uses the action expert alone, preserving real-time, high-frequency control with no added latency. On the VLA-Arena leaderboard(Aug 1, 2026), StellaVLA ranks first with an overall score of 0.63, versus 0.44 and 0.22 for the strong prior models ($π_{0.5}$ and LingBot-VLA), and it further leads on LIBERO with 98.8% average success rate and LIBERO-Plus with 85.1% success rate. Our real-robot benchmark demonstrates that StellaVLA can use both human/robot demos and human-to-robot (XR) demos as in-context structured demonstration to help VLA model adapt to OOD tasks.

Vision-Language-Action (VLA) models can follow instructions and manipulate objects, but their performance often collapses out of distribution (OOD), when the scene, viewpoint, or object differs from training.

We present Ego-OSCAR, an open-hardware, low-cost, head-mounted stereo-inertial capture device for egocentric data collection in the wild. EgoOSCAR pairs a hardware-synchronized global-shutter stereo camera with a 6- axis IMU, an embedded Linux SBC for on-device video encoding, and a realtime microcontroller for user feedback and watchdog functions. The complete bill of materials is under USD 200 per unit, using only commercially available components and 3D-printed parts. Alongside the device, we release a complete software stack (hardware-accelerated recording pipeline, IMU sampling daemon, time-synchronization tooling, and watchdog firmware) and roughly 550 hours of egocentric stereo video per camera with synchronized IMU, collected by a distributed contributor network across everyday indoor environments. The release is annotated rather than raw: free-form action captions cover essentially the entire recorded timeline with an open vocabulary, and per-frame 3D hand reconstructions ship alongside per-session stereo calibration. Ego-OSCAR does not aim to match the per-unit fidelity of research-grade systems such as Project Aria; it aims to be the cheapest defensible substrate for crowdsourced egocentric capture, and to lower the activation energy for any team that wants to collect egocentric data at scale. All hardware designs, software, and the dataset are open-sourced

We present Ego-OSCAR, an open-hardware, low-cost, head-mounted stereo-inertial capture device for egocentric data collection in the wild.

Fine-tuning pre-trained robot policies with reinforcement learning (RL) often inherits the bottlenecks introduced by pre-training with behavioral cloning (BC), which produces narrow action distributions that lack the coverage necessary for downstream exploration. We present a unified framework that enables the exploration necessary to enable efficient robot policy finetuning by bridging BC pre-training and RL fine-tuning. Our pre-training method, Context-Smoothed Pre-training (CSP), injects forward-diffusion noise into policy inputs, creating a continuum between precise imitation and broad action coverage. We then fine-tune pre-trained policies via Timestep-Modulated Reinforcement Learning (TMRL), which trains the agent to dynamically adjust this conditioning during fine-tuning by modulating the diffusion timestep, granting explicit control over exploration. Integrating seamlessly with arbitrary policy inputs, e.g., states, 3D point clouds, or image-based VLA policies, we show that TMRL improves RL fine-tuning sample efficiency. Notably, TMRL enables successful real-world fine-tuning on complex manipulation tasks in under one hour. Videos and code available at https://weirdlabuw.github.io/tmrl/.

Fine-tuning pre-trained robot policies with reinforcement learning (RL) often inherits the bottlenecks introduced by pre-training with behavioral cloning (BC), which produces narrow action distributions that lack the coverage necessary for downstream exploration.

Motion-centric video reasoning is fundamental to interactive applications such as robotic manipulation and autonomous navigation. However, multimodal large language models (MLLMs) typically process videos through sparse uniform sampling to control visual-token and attention costs. This strategy may discard critical transitions between sampled frames, limiting reasoning about object movement, collisions, and causal interactions. To mitigate this issue, we propose Motion-as-Prompt (MaP), a track-guided cross-frame visual prompting framework. MaP recovers dense point trajectories, selects motion-informative frames, and marks the trajectories accumulated between consecutive sampled frames directly onto the visual inputs, making otherwise hidden displacement, direction changes, and interactions observable to frozen MLLMs. Experiments on CLEVRER and Something-Something-v2 show that MaP consistently improves average motion-reasoning accuracy, yielding gains of 4.2% and 8.9% for GPT-5.5, respectively. Notably, these improvements are obtained without degrading non-motion understanding, highlighting the robustness of MaP. These results demonstrate that MaP provides a simple and effective solution for enhancing motion-centric video reasoning without model training or architectural modification. Project page:https://github.com/SunVictor23/MaP.

Motion-centric video reasoning is fundamental to interactive applications such as robotic manipulation and autonomous navigation.

Although mixture-of-experts, MoE, models have been increasingly adopted to scale large language models with moderate computation cost, it remains challenging to deploy MoE inference over resource-constrained and bandwidth-limited edge infrastructures. Existing distributed MoE serving methods mainly rely on exact expert placement, caching, replication, or communication scheduling, while overlooking the functional similarity among experts, which provides an opportunity to reduce cross-server token transmission. Therefore, this paper introduces a similarity-aware expert allocation and distributed deployment framework, dubbed OrderMoE, which aims to accelerate edge MoE inference while balancing inference latency, communication overhead, server workload, and inference quality. OrderMoE first constructs an expert similarity model based on router-induced logits representations and partitions experts in each MoE layer into multiple similarity groups. Then, it develops a similarity-aware expert grouping and deployment strategy to improve local similarity coverage across edge servers. Since reducing remote expert invocation and preserving exact inference quality are conflicting objectives, OrderMoE further designs a quality-aware and trajectory-aware runtime server-expert selection algorithm to decide whether a token should invoke its remote target expert or use a feasible local substitute expert. Experimental results on a real distributed edge testbed show that OrderMoE significantly reduces average latency, tail latency, cross-server traffic, and remote expert invocation ratio, while introducing only small and controllable inference quality degradation.

Although mixture-of-experts, MoE, models have been increasingly adopted to scale large language models with moderate computation cost, it remains challenging to deploy MoE inference over resource-constrained and bandwidth-limited edge infrastructures.

Exploration is essential to RL since a policy cannot improve by repeatedly sampling the behaviors it already prefers. Standard methods inject stochasticity in the action space, but such jitter only yields rollouts close to the original. Escaping a weak policy often requires global perturbations that action noise cannot produce. Large language models (LLMs) and vision-language-action (VLA) models offer a pathway: they condition the policy on a natural language prompt, and since the rollout follows from it, modifying the prompt induces global changes. The challenge is finding prompts that induce useful global changes. With a weak policy that rarely succeeds, reward is too sparse to select on. Our idea is to refine prompts from the rollouts themselves: a vision-language model (VLM) reasons over the rollout video, diagnoses how the policy responded, and rewrites the prompt to elicit better behavior next time. This procedure resembles posterior sampling, a classical RL exploration framework, at the level of prompts: the VLM maintains an implicit distribution over useful prompts and updates it from observed rollouts. We call this strategy Prompt-Driven Exploration (PDE). Across manipulation and reasoning tasks, PDE enables RL to learn successful policies even from zero-reward starts, and improves sample efficiency more broadly. Our website is available at https://xinyunsunshine.github.io/prompt-rl.

Exploration is essential to RL since a policy cannot improve by repeatedly sampling the behaviors it already prefers.

Natural-language-based scenario generation offers an intuitive means of describing rare and complex driving interactions, yet it is still uncertain whether training with language-structured data leads to truly adaptive control policies. We propose Language-Structured Relational Q-Learning, instantiated through an Ego-Centric Relational Q-Network (ERQ-Net), which jointly learns inter-vehicle relevance and action values from dynamic traffic graphs. Language descriptions define surrounding-vehicle behaviours during training, while prompts and semantic actor roles are hidden from the policy. ERQ-Net must therefore infer threat relevance solely from observable kinematics and interactions. Across 2,500 safety-critical scenarios, language-structured training improves test success from 49-52% to 55-58% and increases adversary-focused attention from 1.2x to 2.1x, demonstrating emergent threat awareness. However, this representational gain does not consistently translate into adaptive control: trained policies perform similarly to the best constant action, while a portfolio of simple policies solves 76% of scenarios. We formalise this discrepancy as a recognition-control gap and show that reward reweighting and margin shaping do not eliminate the resulting policy collapse. Evaluations of realism, criticality, semantic accuracy, and transfer of state-interface representations to CARLA further highlight both the strengths and the constraints of language-structured relational policy learning in safety-critical driving scenarios.

Natural-language-based scenario generation offers an intuitive means of describing rare and complex driving interactions, yet it is still uncertain whether training with language-structured data leads to truly adaptive control policies.

IoT firmware vulnerability detection remains challenging due to heterogeneous firmware ecosystems, resource-constrained platforms, and limitations in existing benchmarks. Many datasets are synthetic or general-purpose and lack human-verified, contamination-screened annotations, limiting evidence on cross-corpus generalization across training sources, model architectures, and curriculum strategies. To address this gap, this paper introduces IoTVulBench, a human-verified benchmark for cross-corpus firmware vulnerability detection. IoTVulBench-Core was constructed from GitHub repositories, validated by three expert reviewers, and evaluated on a contamination-screened held-out target across five model architectures, two tuning methods, and three curriculum strategies, with ensemble, distillation, and robustness analyses. Models trained on IoTVulBench achieved the highest MCC among matched single-source datasets, reaching 0.58 versus 0.44 for PrimeVul and 0.39 for D2A. Staged curriculum learning increased MCC to 0.69, while a diversity-optimized ensemble achieved 0.73, improving by 0.42 MCC over the strongest reference comparator, a static analyzer at 0.31, and by 0.29 over PrimeVul. At a 0.5% false-positive rate, the model missed only 21% of vulnerabilities, compared with 71% for the strongest comparator. It retained 86% of its performance under identifier renaming and demonstrated strong calibration and largely faithful explanations. These findings indicate that domain-matched training data and curriculum design, rather than model scale alone, are key drivers of generalization in firmware vulnerability detection. The results provide a benchmark for future research and deployment-ready configurations for practical IoT security applications.

IoT firmware vulnerability detection remains challenging due to heterogeneous firmware ecosystems, resource-constrained platforms, and limitations in existing benchmarks.

Extracting quantum information from a quantum state is a fundamental task of quantum computation, often requiring the estimation of many non-commuting observables under a finite measurement budget. For both near-term and early fault-tolerant settings, the measurement protocol must balance statistical efficiency against implementation resources such as circuit depth, connectivity, and entangling-gate count. Many existing strategies focus on two extremes: hardware-friendly product measurements with high sampling cost, and fully commuting measurements with deep circuits. Here we recast resource-constrained measurement design as a generative learning problem. We introduce FlowMeas, which uses a generative flow network to directly sample finite ensembles of shallow Clifford measurement circuits subject to a prescribed shot budget and hardware constraints. At zero entangling depth, FlowMeas learns qubit-wise commuting measurement schedules and already matches or improves leading product-measurement methods on nearly all molecular benchmarks. Allowing one or two entangling gate layers yields further reductions in energy estimation error of up to $27\%$ relative to the strongest state-independent product-measurement baseline. The learned policy can also be reused across related Hamiltonians, substantially accelerating retraining along a molecular potential-energy surface. We further obtain results for molecular Hamiltonians with up to 20 qubits and apply the framework to a compactly encoded 54-qubit interacting fermionic model, extending the demonstrated scale beyond prior molecular benchmarks. These results establish generative learning as a flexible and unified framework for quantum measurement design under practical resource constraints.

Extracting quantum information from a quantum state is a fundamental task of quantum computation, often requiring the estimation of many non-commuting observables under a finite measurement budget.

Branch mispredictions and data-cache misses are usually evaluated as separate bottlenecks: studies report perfect-branch or perfect-cache speedups as isolated upper bounds and often treat their product as the joint ceiling. In irregular workloads, however, hard-to-predict branches and cache-missing loads often occur in the same hot loops. Removing one penalty can expose the other: faster memory reaches mispredicted branches sooner, while better branch prediction leaves more long-latency loads in the out-of-order window. We call this interaction symbiotic stall latency (SSL). This paper quantifies when isolated ceilings fail using joint speedup synergy (JSS), the observed joint perfect-branch/perfect-cache speedup divided by the product of the isolated speedups. Values above one mean independent-ceiling analysis understates attainable gain. Across 53 simulated workloads, 70% show measurable coupling (JSS > 1), though many are near unity, especially in lower-pressure cases. With a conservative threshold, 40% exceed the independence product by more than 6%, and kernels with SSO > 20 show JSS from 1.23 to 3.29. We introduce symbiotic stall opportunity (SSO), a lightweight MPKI-based screen for workloads that merit full joint simulation. We map high-SSO workloads to four recurring software patterns: neighbor access, hash lookup, linked-structure traversal, and data-dependent modification. We connect SSL to reorder-buffer occupancy, squash rate, and commit starvation under isolated perfect modes. The resulting methodology is simple: use SSO to screen, JSS to validate, and report conditional branch-after-cache and cache-after-branch gains when evaluating branch predictors, prefetchers, caches, or coupled branch/memory mechanisms. Our contribution is a measurement framework showing when isolated perfect modes are adequate and when they understate joint performance headroom.

Branch mispredictions and data-cache misses are usually evaluated as separate bottlenecks: studies report perfect-branch or perfect-cache speedups as isolated upper bounds and often treat their product as the joint ceiling.

A central goal in robot learning is to move beyond task-specific human data collection toward robots that improve through autonomous interaction. Yet fully autonomous learning remains difficult with current policies: sparse rewards and weak zero-shot exploration make it unlikely that a robot will discover successful behavior from scratch. We study minimal-data adaptation, a regime in which a pre-trained robot policy must learn a new task from as little as one demonstration followed by autonomous online interaction. This setting serves as the closest tractable proxy for fully autonomous improvement, allowing us to study whether minimal human guidance can bootstrap autonomous learning and what algorithmic ingredients make it feasible. We build MiDAS, a simple offline-to-online RL recipe that first anchors a pre-trained VLA to the target task with behavior cloning on single/few demonstrations, then improves it through value-based online RL on a residual policy parameterization. Across LIBERO and RoboCasa, MiDAS recovers strong task performance from as little as one demonstration, substantially outperforming baselines and generalizing beyond demonstrated conditions. We further evaluate MiDAS on a bimanual YAM platform. Starting from a fragile low-success policy obtained from a single demonstration, MiDAS improves its robustness and learns new successful behaviors over ~6 hours of online interaction. To the best of our knowledge, this is the first demonstration of reliable robot policy adaptation from a single task demonstration.

A central goal in robot learning is to move beyond task-specific human data collection toward robots that improve through autonomous interaction.

While learned robotic policies hold promise for advancing generalizable manipulation, their practical deployment is often hindered by suboptimal execution speeds. Imitation learning policies are inherently limited by hardware constraints and the speed of the operator during data collection. In addition, there are no established methods for accelerating policies learned via imitation, and the empirical relationship between execution speed and task success remains underexplored. To address these issues, we introduce SpeedTuning, a reinforcement learning framework specifically designed to enhance the speed of manipulation policies. SpeedTuning learns to predict the optimal execution speed for actions, thereby complementing a base policy without necessitating additional data collection. We provide empirical evidence that SpeedTuning achieves substantial improvements in execution speed, exceeding 2.4x speed-up, while preserving an adequate success rate compared to both the original task policy and straightforward speed-up methods such as linear interpolation at a fixed speed. We evaluate our approach across a diverse set of dynamic and precise tasks, including pouring, throwing, and picking, demonstrating its effectiveness and robustness in enhancing real-world robotic manipulation. Videos and code are available at https://daivdyuan.github.io/speed-tuning/

While learned robotic policies hold promise for advancing generalizable manipulation, their practical deployment is often hindered by suboptimal execution speeds.

Deploying Vision-Language Models (VLMs) on edge devices remains challenging due to their substantial computational and memory demands, which exceed the capabilities of resource-constrained embedded platforms. Conversely, fully offloading inference to the cloud is often impractical in bandwidth-limited environments, where transmitting raw visual data introduces substantial latency overhead. While recent edge-cloud collaborative architectures attempt to partition VLM workloads across devices, they typically rely on transmitting fixed-size representations, lacking adaptability to dynamic network conditions and failing to fully exploit semantic redundancy. In this paper, we propose a progressive semantic communication framework for edge-cloud VLM inference, using a Meta AutoEncoder that compresses visual tokens into adaptive, progressively refinable representations, enabling plug-and-play deployment with off-the-shelf VLMs without additional fine-tuning. This design allows flexible transmission at different information levels, providing a controllable trade-off between communication cost and semantic fidelity. We implement a full end-to-end edge-cloud system comprising an embedded NXP i.MX95 platform and a GPU server, communicating over bandwidth-constrained networks. Experimental results show that, at 1 Mbps uplink, the proposed progressive scheme significantly reduces network latency compared to full-edge and full-cloud solutions, while maintaining high semantic consistency even under high compression. The implementation code will be released upon publication at https://github.com/open-ep/ProSemComVLM.

Deploying Vision-Language Models (VLMs) on edge devices remains challenging due to their substantial computational and memory demands, which exceed the capabilities of resource-constrained embedded platforms.

Assessing catheter and tube placement on chest X-rays is safety-critical yet tedious and error-prone. Current deep learning methods either classify placement globally -- losing track of which device is where -- or segment all devices into a single mask, making per-device assessment impossible when catheters overlap. We introduce UCompCXR, a compositional framework that detects local catheter fragments, associates them into device instances via graph-based clustering, fuses per-fragment tip predictions through precision-weighted Gaussian estimation, and classifies placement per device. On the RANZCR CLiP dataset (30,083 images, 5-fold patient-level CV with bootstrap CIs), UCompCXR detects 26% more devices than a strong multi-task baseline sharing the same MobileNetV3 backbone, with 75% fewer false positives and well-calibrated tip uncertainty (95% coverage = 0.948). The aggregate tip error rises -- but only because the model finds devices the baseline misses entirely, especially nasogastric tubes. On matched devices, catastrophic localization failures drop substantially. At 2.27M parameters in a single forward pass, the model is deployable on resource-constrained clinical hardware.

Assessing catheter and tube placement on chest X-rays is safety-critical yet tedious and error-prone.