Localizing object boundaries is important for embodied visual systems, since small boundary errors can change the inferred geometry of physical surfaces. We study whether deep stereo models produce boundaries consistent with simple shape priors when local image and disparity cues are weak. We create a synthetic stereo dataset consisting of objects drawn from a highly constrained family of shapes, with randomized depth, slant, color and texture. Despite the simplicity of this shape space, we find that several state-of-the-art stereo models retain systematic boundary errors, even when finetuned or trained from scratch on the dataset. We then introduce a simple generative stereo model that jointly predicts disparity and a distance field representing object boundaries. The model improves boundary localization and produces contours that adhere more closely to the underlying shape space. These results expose limitations in the boundary predictions of current stereo architectures when local evidence is weak, and motivate further study of whether generative models can better combine regional grouping evidence with global geometric structure.
https://arxiv.org/abs/2611.25113
Recent advances in world models are opening new possibilities for fine-grained embodied tasks, with hand-object interaction offering a particularly demanding setting. Modeling these interactions involves capturing how hand actions change objects and how the resulting states persist over time. Yet systematic evaluation tailored to these hand-centric capabilities remains underdeveloped. To address this gap, we introduce HandWorldBench, a unified benchmark that systematically evaluates hand-centric world models across human hands, simulated dexterous hands, and real dexterous hands along six dimensions: visual quality, hand fidelity, appearance consistency, instruction following, physical consistency, and temporal coherence. To construct the benchmark, we develop a data curation pipeline that extracts hand-object interaction clips and screens them for quality, yielding 20,500 high-quality egocentric clips from over 600 hours of footage. We then manually select 500 clips covering seven task families for evaluation. Benchmarking eleven models under a unified protocol reveals that stronger hand-pose control does not consistently translate into coherent object-state evolution and persistence or robust cross-embodiment performance, with several specialized hand-control models underperforming general-purpose text/image-to-video generators on dexterous-hand tasks.
https://arxiv.org/abs/2611.24731
Vision-language-action policies control embodied agents through a perception-action loop grounded in physical interaction, where adversarial patches can corrupt action predictions and thereby alter subsequent observations. To mitigate these threats, many existing defenses rely on policy adaptation or auxiliary models, but require additional training or inference. We introduce RECLAIM, a training-free, inference-time defense that uses a frozen VLA's own attention response to local visual interventions to guide purification during closed-loop embodied manipulation. Our key insight is that attention changes induced by local edits provide feedback for choosing where and how to purify as the robot interacts with its environment. RECLAIM combines this feedback with robot geometry to refine patch localization and select local treatments, then revalidates them as the agent's embodied actions change the observed scene. We evaluate RECLAIM against patches generated by four adversarial attack methods across multiple manipulation task suites under frozen VLA control. Under adversarial patch attacks, RECLAIM succeeds in at least one rollout on 95% of the evaluated tasks and raises rollout success on LIBERO-Object from 0% without defense to 70.0%. Its mean decision time is only 548.9 ms per control step. These results highlight policy-guided visual purification as an effective approach to defending embodied agents against adversarial patches and recovering closed-loop task execution.
https://arxiv.org/abs/2611.24494
Embodied AI systems, such as drones and autonomous vehicles, improve navigation by enhancing perception in environments containing unseen objects. While Visual Language Models (VLMs) offer strong image understanding, their real-world application is underexplored. For effective deployment, open-vocabulary semantic segmentation (OVSS) models, leveraging models like CLIP to assign pixel-level labels from text, must be robust to outdoor conditions, adverse weather, and varying viewpoints. Despite their potential, these models encounter significant challenges in cross-domain generalization, hindering their practical use in embodied AI applications. In this work, we present Pixel2Perspective, a benchmark for multi-angle object localization and open-vocabulary segmentation across both aerial and ground perspectives. This benchmark supports thorough evaluation across different viewpoints and sensor modalities. We assess foundational models like CLIP and advanced OVSS models using a curated dataset that spans varied outdoor conditions, sim-to-real scenarios, and multiple sensors, including infrared, to support the development of robust and generalizable perception systems. We also present a novel framework, Multi-Source RGB-IR Uncertainty Quantification (MSRIQ), which jointly models aleatoric and epistemic uncertainties across RGB and infrared imagery. This approach supports robust evaluation of open-vocabulary segmentation models across diverse environmental conditions. Our work sets a robustness-driven benchmark, providing key insights and laying the foundation for future research in open-world embodied AI perception. The benchmark and project page are available at this https URL
https://arxiv.org/abs/2611.24327
Large-scale robotic simulation requires diverse deformable objects with executable physical properties and realistic interaction behaviors. However, existing 3D asset collections primarily emphasize visual diversity, while scalable physical construction and verification remain challenging. We introduce RFW-S, a large-scale dataset of 83,365 simulation-verified assets spanning 485 everyday categories across five structural families: soft solids, cloth sheets, shells, garments, and bags. We develop an automated pipeline integrating reference-guided 3D generation, structure-aware physical modeling, and LLM-assisted parameter estimation and calibration. To ensure physical reliability, we introduce an independent simulation verification framework that evaluates object behaviors under interactions distinct from those used for parameter calibration. Extensive experiments demonstrate the effectiveness of independent verification and characterize deformable responses under different physical configurations. We further evaluate robotic grasping in terms of both task success and object integrity, and demonstrate the physical realism of our simulated objects through sim-to-real replay of fragile-object manipulation. RFW-S provides a scalable foundation for physically grounded simulation and embodied robotic learning.
https://arxiv.org/abs/2611.23056
Embodied models depend on real interaction data, yet collecting it on physical robots is slow and labor-intensive. Recent ego-to-robot pipelines retarget first-person view human videos to robot data via transferring hand and arm motion. They judge fidelity mainly by visual appearance. However, trajectories often fail in terms of physics: fingers hover beside objects, and grippers close on empty space. We argue that how the object moves and the contacts that drive the motion count. Thus, we present Ego2X, an object-centric framework that converts monocular egocentric video into physically executable robot data across dexterous hands and parallel grippers. Ego2X centers on CORE (Contact-centric Object-interaction REpresentation), which records each contact on the object by its surface anchor, as well as relative force, grasp mode, and interaction state. Throughout the pipeline, CORE corrects hand-object drift via object-relative constraints, steers inverse kinematics toward contacts the robot must realize, and preserves held objects during arm erasure and robot compositing. On HOT3D, contact-driven correction reduces in-contact hand depth error by 33.7%, while on EgoDex, contact-driven retargeting realizes 62.9% of annotated contacts, surpassing the official HuRo pipeline (27.9%). In RoboTwin 2.0, post-training X-VLA on Ego2X episodes alone, without simulator demonstrations, raises task success rate from 40.5% zero-shot to 67.0% across 200 paired scenes. We will publicly release our conversion pipeline and CORE-annotated dataset for dexterous hands and parallel grippers.
https://arxiv.org/abs/2611.24138
Long-horizon tasks in embodied and web environments involve extended decision sequences and delayed feedback, making accurate planning crucial for success. World models tackle this by learning latent dynamics for imagination-based planning. However, existing approaches merely rely on dense pixel-level observations, which forces the model to absorb redundant information while overlooking relations among task-relevant entities. Consequently, small errors may accumulate over long imagined rollouts and gradually corrupt the predictions. In this paper, we explore world models that imagine over sparse state abstractions, which poses two key challenges: 1) how to remove redundant information and form a compact state abstraction; 2) how to incorporate the relational structure to learn reliable long-horizon dynamics. To address these challenges, we propose Embodied Graph, a relational state abstraction that supports structured dynamics modeling and long-horizon imagination, and the Relational World Model (RWM), which performs imagination directly over graph states. RWM consists of (i) a Task-Aware Embodied Graph Constructor that filters redundant details and maintains graph memory through online updates, and (ii) an Embodied Graph Recurrent State Space Model that learns node-wise latent dynamics while capturing inter-node dependencies. The imagined graph states provide predictions of action consequences to guide decision-making. Experiments on Minecraft and Crafter demonstrate the effectiveness of RWM in embodied environments. Notably, in Minecraft, RWM is the only method that obtains diamonds within the given budget. Experiments on web interaction further illustrate the applicability of relational state prediction to the decision making process of web agents.
https://arxiv.org/abs/2611.24115
Embodied robots increasingly rely on large language models (LLMs) to interpret user intent and coordinate with robot-specific action models for physical interaction. This multi-model pipeline creates a critical semantic boundary across mind, model collaboration, and motion: semantic consistency may break between what the large model understands, how the small action model translates that understanding into executable commands, and what physical actions the robot ultimately performs. As a security consequence, commercial robots may execute unintended physical actions even when user inputs contain no explicit action commands. To measure this problem, we develop TARS, a black-box framework that explores the command-reasoning-action pipeline and automatically discovers and validates unintended physical behaviors. We evaluate TARS on four commercial robots across 44 experimental settings, conducting 133,400 executions. TARS uncovers widespread cross-model action misalignment and achieves up to 94% verified action coverage.
https://arxiv.org/abs/2611.24100
Wearable ego-centric images are now in high demand for scenarios ranging from daily smart glass usage to embodied intelligence. But the image quality is far behind smart phone due to the lacking paired high-quality reference images and the dedicated enhancement systems. To solve this, a customized degradation pipeline is designed to generate paired samples matching wearable camera images, boosting the upper limit of enhancement performance. Besides a two-stage enhancement framework is further built: first, tuning an efficient model on the paired dataset to enhance real wearable images; second, using hyperspectral data to refine color temperature for better quality. Moreover we present HHW-Ego, a multi-source paired dataset integrating with Hyperspectral, High-dynamic range and Wearable Ego-centric data, which includes 5,000 image groups spanning indoor/outdoor and day/night scenes. On held-out paired data, Stage-1 fine-tuning improves PSNR, SSIM, Lab distance, and LPIPS for Real-ESRGAN and PFT-SR. On real wearable images, no-reference metrics, human preference, gamut analysis, and visual comparisons characterize the effects of Stage 2.
https://arxiv.org/abs/2611.24010
Vision-language agents that browse websites, operate phones and act in homes reason before every action, so every step waits for a large model, a latency and cost that stand in the way of deployment. Step-level speculation lets a small model draft each step and calls the large model only on rejected drafts, but its outcome rests on the judge, and a judge that reads only a step's text cannot tell whether the step fits the screen it acts on. We find that a few samples of a small model can contain a correct step about as often as the large model gets the step right, and not always on the same steps: checked against the screen, small drafts can be not only cheaper but better. We present Vanguard, a training-free judge that leaves the draft and target models unchanged: a general decision model, with the screenshot in view, gives each candidate step a probability of being right in one forward pass; the best candidate is kept if it clears a threshold, and otherwise the large model writes the step. On web, mobile and embodied agents, a 2B draft judged by Vanguard is at least as accurate as its 8B target and significantly more accurate on Mind2Web and EB-ALFRED, while the target writes only a minority of the steps; it completes far more closed-loop episodes, and with a 27B target from another family it is both faster and more accurate. Vanguard is the most accurate judge we compare on the agent tasks; it relies on the screenshot on web and mobile tasks, its calibrated threshold transfers to unseen websites, and across model families its gain over the target grows with the draft's pass@4 relative to the target's accuracy. Vanguard composes with token-level speculative decoding and matches trained process reward models on mathematics, turning speculation from a way to save the large model's work into one that can improve on it.
https://arxiv.org/abs/2611.23917
To advance the development of embodied general navigation, zero-shot open-vocabulary multi-object navigation has attracted increasing attention, aiming to enable robots to efficiently search multiple objects in unseen environments without training. For this task, the key lies in improving navigation policies from earlier searches to locate subsequent goals more efficiently. Recent methods automate this process through trajectory reflection and rule revision, enabling policy self-improvement. However, the mechanisms for diagnosing problems and generating revisions are typically designed by humans and remain fixed across iterations. Consequently, the system's capacity for continued self-improvement remains constrained, even as its navigation policy evolves. To this end, we propose RPSI-Nav, a Recursive Policy Self-Improvement framework for zero-shot multi-object navigation that revises both the navigation program and its revision mechanism. RPSI-Nav diagnoses search behavior, requests diagnostic views when existing observations cannot distinguish competing causes, and generates targeted navigation revisions. Feedback from subsequent goal searches is linked to the rules that produced these revisions, guiding changes to diagnosis, observation acquisition, and revision planning. The updated improver governs subsequent navigation improvements, forming a recursive loop in which RPSI-Nav revises its own improvement process. Across goals, the robot conditionally inherits accepted program updates and spatial memory, enabling search experience to improve both future navigation and adaptation. Benchmark experiments and qualitative real-robot evaluations validate RPSI-Nav's effectiveness.
https://arxiv.org/abs/2611.23774
Dynamic spatial reasoning from monocular video enables vision-language models (VLMs) to understand how 3D environments evolve over time, supporting physical world understanding and embodied intelligence. Existing efforts primarily address observed-dynamics understanding, whereas agents in dynamic environments also require future-dynamics prediction beyond the observation boundary. Moreover, external geometric processing and textual descriptions can increase inference complexity or lose information when representing continuous dynamics. These limitations motivate a shared internal state that preserves observed dynamics and remains predictive beyond the input video. Accordingly, we present Pros4D, a framework for modeling evolving 4D worlds in latent space that unifies retrospective understanding of observed-dynamics with prospective reasoning for future-dynamics prediction. Specifically, Pros4D-Engine derives geometry-grounded supervision and constructs paired examples for both tasks, while Pros4D-Bench evaluates them across eight spatial task categories spanning dynamic cameras, objects, and scenes. At the model level, Pros4D compresses observed dynamics into latent observed states and autoregressively rolls them forward into latent future states. 4D Latent Alignment (4DLA) grounds these states through complementary visual-semantic and geometric supervision, while joint language supervision connects them to task reasoning. Experiments across benchmarks show consistent improvements over strong baselines in both capabilities.
https://arxiv.org/abs/2611.19803
Animals generate behavior through continuous interactions between their neural circuits, bodies, and environment. Embodied simulations study these interactions by testing neural circuit models in biologically realistic settings. Recent efforts have simulated diverse animals, and emerging work on biologically grounded neural architectures has begun to integrate circuit connectivity and single-unit dynamics from experimental data into artificial neural networks, introducing stronger inductive biases and closer mechanistic alignment with biology. How can we develop embodied simulations that enable the study of increasingly complex bodies and neural circuits? In this work, we address the challenge through coordinated advances in the simulated body, neural architecture, and training infrastructure. We introduce Rat v2, an efficient and flexible biomechanical body for simulating rat behavior. To control this body, we develop Rat NCAP, a biologically grounded neural architecture for locomotion based on mammalian spinal circuits that incorporates cell-type-specific connectivity and a novel flexor-extensor neuromechanical interface. To optimize circuit parameters, we develop GPU-accelerated simulation and training infrastructure. The model learns to locomote at different speeds and transition between gaits without imitation learning, producing locomotion statistics, kinematics, and cell-type-specific neural activity consistent with experimental data. Analyses reveal that the neuromechanical interface substantially affects trainability and gait quality, and in silico cell-type ablations reproduce known gait phenotypes and predict gait changes under graded perturbations. Together, these findings support the view that embodied simulations at an intermediate level of abstraction can simultaneously reproduce complex behavior and yield mechanistic insights into the neural circuits that generate it.
https://arxiv.org/abs/2611.23364
Large language models (LLMs) in embodied robots create security dependencies beyond inference. We investigate a network-exposed configuration attack on the Pollen Robotics Reachy Mini Lite, allowing an unauthenticated attacker to replace personality settings and the cloud LLM's system prompt through a management interface, without adversarial conversational input. Eleven prompt-injection attempts across five categories were reported as resisted, whereas both configuration-layer attack paths produced observable behavioural changes on physical hardware. We demonstrate local and network-adjacent access, mapping the weakness to CWE-306 (Missing Authentication for a Critical Function) and CWE-1327 (Binding to an Unrestricted IP Address), with a reported CVSS v3.1 score of 6.5 (Medium). A configuration endpoint also accepts cross-origin requests without a CORS preflight (CWE-352; 6.5 Medium), while the persistent-memory sink stores content without enforcing its stated restrictions. A source review of 59 public repositories identifies missing authentication in 10 of 11 source-verifiable genuine applications, or 11 of 11 including a framework-default case. These findings indicate recurrence within the sampled repositories, not prevalence in commercial products. We characterise a deployment gap between developer-oriented network assumptions and consumer-facing embodied AI security: inference-layer refusal cannot prevent unauthorised changes to model instructions. Potential consequences include surveillance through onboard cameras and microphones and social engineering of nearby users. Proof-of-concept code, evidence and the survey are available at this https URL.
https://arxiv.org/abs/2611.22761
An embodied speaker's expressiveness depends not only on what is said, but also on how the voice sounds and how the body moves. However, existing joint speech-motion generation models are largely driven implicitly by transcript semantics or intermediate speech representations, offering little explicit control over motion hierarchy, body parts, or vocal style. To bridge this gap, we construct MotionSpeech, a structured benchmark for controllable joint speech-motion generation that includes transcripts, voice-style descriptions, and motion conditions at two hierarchical levels. Building on this benchmark, we propose DuoFlow, a dual-stream latent flow transformer that jointly denoises speech and motion under a shared noise level and a unified second-based temporal coordinate. Modality-specific residual connectors inject structured conditions into the corresponding streams, layer-wise bidirectional cross-modal attention coordinates their generation, and a monotonic phoneme pathway combines normalized progress anchors with CTC supervision to strengthen speech temporal modeling. DuoFlow provides a unified interface for fine-grained motion and vocal-style control and demonstrates the ability to synthesize high-quality, synchronized, and instruction-consistent speech-motion sequences. Project page, this https URL.
https://arxiv.org/abs/2611.21530
JEPA-style latent world models let embodied agents plan toward new goals with model predictive control, but they are still too slow for real-life embodied control: sampling-based MPC repeats predictor inference tens of thousands of times per plan, and even on high-end GPUs one planning step can take tens of seconds. We observe substantial latent locality in this repeated-inference workload: the predictor maps close actions to close latents, and inside a quantized predictor this closeness takes a form that hardware can exploit, differential bit-level sparsity. Unlike an input stream, whose temporal similarity is given, a planner itself decides which candidates it evaluates together, so the planner can choose its own references and, with reference forest sampling, manufacture more sparsity: it couples the sampled action sequences of each group to a shared root, with competitive planning quality. We build the accelerator on the insights of the rich line of bit-level sparsity accelerators and design it and the planner tightly coupled as a single co-design: the planner increases sparsity and exposes structure for hardware scheduling, and the bit-exact hardware effectively uses the sparsity that the planner creates. On three JEPA world models, our co-design is 1.28-1.80x faster than a dense accelerator and uses 38-52% less energy, and it is 1.22-1.58x faster than prior bit-sparse design and uses 16-36% less energy, with the same area. To our knowledge, our planner is the first planner that manufactures differential bit-level sparsity across its candidates and exposes its structure for hardware scheduling, and our co-design is the first algorithm-hardware co-design for JEPA-style world-model planning.
https://arxiv.org/abs/2611.22483
Changing the holding reference during a fixed-duration wait can reverse the ranking of visual recovery methods. We introduce AuditEmbodied, a paired execution and trajectory evaluation framework that independently varies holding references and waiting duration while keeping recovery executors frozen. A 960-execution factorial on 20 initialization seeds compares CAD-feedback and landmark-based recovery from identical post-wait physical states within each holding and waiting condition. Protected completion requires successful recovery under the declared execution constraints. At a 1 s wait with registered wrist references, their protected completion counts change from 6/40 versus 19/40 under registered holding to 34/40 versus 22/40 under position-only holding. Trajectory analysis localizes the main separation to grasp formation: small observation-time target residuals precede large tracking errors. Joint diagnostics and controlled interventions characterize persistent tracking constraints. On fresh seeds, gradual position reanchoring improves CAD recovery's long-wait completion by 42.5 percentage points and reduces mean complete recovery time by 6.06 s relative to registered holding. Additional evaluations with pretrained policies, handover, and a mass-matched geometry characterize holding-reference sensitivity across executors and physical regimes. These findings establish holding reference as an explicit evaluation condition for comparing visual recovery performance.
https://arxiv.org/abs/2611.22368
Vision-language-action (VLA) models drive the robot directly from camera views, yet a VLA alone shows three typical failures. It walks past the goal, stops before reaching it, and rarely recovers after going off track. Pairing a VLA with a large language model (LLM) planner fixes these failures, but queries the LLM throughout every episode, adding token costs and latency. We introduce Harness-to-Executor (H2E), a task-level distillation method that pairs an LLM planner and a frozen VLA as a heterogeneous training agent, keeps only sessions that reach the goal, and then trains a VLA alone on them so no LLM is used at test time. In these sessions, the planner gives sub-instructions and moves the robot with its own tools, detours are common, and the planner decides when to stop. H2E therefore labels every motion as a VLA action for the original task, whoever made it. It masks the loss on steps moving away from the goal and assigns end labels to states near it, so the VLA learns where the task ends. H2E sets a new state of the art on REVERIE-CE and NavRAG-CE. It raises an 8B VLA's success rate by 15.6% and 11.1% respectively, surpassing a method trained on 10M samples. The distilled VLA exceeds its agent on unseen episodes with zero LLM tokens and runs 3.7 times faster.
https://arxiv.org/abs/2611.16246
What protects an LLM team when one teammate quietly goes wrong? Most multi-agent LLM work assumes cooperative partners and inspects chat between turns before acting. Real-time embodied cooperation does not allow this pause: chat runs alongside action, so a teammate whose words stay cooperative while behavior diverges can cause damage faster than the team can name what is happening. We study this case by injecting three theory-grounded misaligned agents into a custom four-player Overcooked environment, each stressing a different joint-action layer (speech-action mismatch, role invasion, and aggregate-progress disruption), and decompose robustness into three sequential gates of detection, correction, and calibration through a 3x4x2 design (n=720 episodes). Nearly all trolls are eventually named, but recovery follows only when attribution and situational diagnosis appear in the same utterance; direct source-attribution prompting improves the hardest case only by degrading the troll-free baseline by 37.5%. Robustness, then, depends less on detection alone than on whether team chat combines attribution with diagnosis in a single utterance, suggesting that sensitivity to misalignment should be targeted rather than uniformly raised.
https://arxiv.org/abs/2611.22342
Interactive video world models serve as simulators for embodied agents: from a starting image and an action trajectory, they generate the views the agent would see. The starting image fixes how high the camera is, for example 0.15 m for a floor robot or 2.2 m for an overhead camera, and a faithful simulator must keep this height as the camera moves. We show that six recent open-weight models fail to do so: as motion replaces the starting view with generated content, they move the camera toward human eye level, a failure we call camera-height regression. Our benchmark, EyeHeight, pairs six rendered elevations (0.15 m to 2.2 m) and four calibrated real-world rig heights with scale-free multi-view height estimation. All six models lose their initial height during a $180^\circ$ turn-and-walk rollout: initialized at 0.15 m, only 0-20% of clips end within $1.3\times$ of the starting height as inferred viewpoints converge to human heights of 1.1 m to 1.5 m, whereas ground-truth renders remain within a few percent. In Matrix-Game models, homography-tracked floor motion stays consistent with constant elevation during rotation even as newly outpainted room geometry yields a higher reconstructed viewpoint. Regression grows with rotation angle, begins while the start frame remains in context, and largely reverses in SANA-WM when a $360^\circ$ turn restores the initial view. We trace this pull to human-centric video priors in which floor-level views are nearly absent: in SANA-WM, textual height cues and dense floor-disparity maps do not prevent it, whereas classifier-free text guidance during the first half of denoising strengthens it. Combining unguided sampling with height-balanced fine-tuning recovers more retention but costs 5 MUSIQ points, a loss fixed during those same early steps. Viewpoint drift alone shifts the paths predicted by visual navigation policies.
https://arxiv.org/abs/2611.21782