Embodied Reasoning constitutes a fundamental capability of embodied intelligence, serving as the basis for autonomous perception, reasoning, and interaction within physical environments. Recent studies have shifted the paradigm of embodied reasoning from static perception toward dynamic exploration, where agents acquire task-relevant information through interactions with the environment. However, existing active reasoning approaches generally generate exploration trajectories incrementally without long-horizon planning. Even recently emerged test-time scaling frameworks often resort to myopic, single-step lookaheads, which struggle to resolve the delayed feedback inherent in complex, occluded spatial environments. To address this limitation, we propose ParallelWorld, a multi-horizon test-time scaling framework for embodied reasoning. Instead of greedy, single-step trials, ParallelWorld empowers agents to simulate and evaluate multi-step future trajectories in parallel before committing to an action. Specifically, we introduce a verifier-guided tree-search paradigm. Starting from the current state, ParallelWorld branches into multiple parallel trajectories and rolls them out continuously across a multi-step horizon. At each simulation step, a verifier agent evaluates the intermediate state transitions, dynamically pruning unpromising branches and prioritizing paths with the highest information gain. Once the multi-step prospective simulation is complete, the agent synthesizes the long-horizon outcomes to commit to the optimal action sequence. Finally, an answer agent performs reasoning over the selected trajectory to produce the final reasoning. Extensive experiments on ESI-Bench demonstrate that ParallelWorld consistently improves active perception and reasoning performance.
Embodied agents must often identify and interact with objects based on their function rather than their identity, requiring them to actively acquire observations that reveal discriminative functional evidence. Existing affordance grounding methods operate from fixed viewpoints and lack mechanisms for deciding where to look when functional cues are occluded or incomplete. We introduce Active Functional Affordance Grounding, a new task in which an agent sequentially explores a scene to identify and spatially ground an object satisfying a functional query. To address this problem, we propose FUSE, an adaptive semantic-geometric evidence acquisition framework that combines explicit uncertainty-driven exploration with a learned amortized planner to efficiently select informative viewpoints. We further introduce a Habitat-based benchmark for evaluating active functional grounding. Experiments show that FUSE achieves the highest observed non-oracle grounding performance while reducing computation by 1.33x relative to fully explicit exploration, and remains effective across multiple affordance knowledge sources.
Natural language provides robots with a flexible task interface, but target ambiguity in embodied environments arises not only from user intent; it can also result from missing taskrelevant physical evidence in the current observation. Existing interactive disambiguation methods primarily obtain additional information by asking the user, whereas occlusion, restricted viewpoints, unreadable text, and unobserved targets require the robot to actively change its observation. We propose an active-perception framework for embodied target disambiguation that uses active observation as the backbone for information acquisition and uses a vision-language model to decide, on the basis of accumulated visual evidence and interaction information, whether to continue observing, request clarification, or complete target selection. Active observation can both directly recover missing discriminative evidence and reveal object names, labels, and semantic attributes, thereby improving user clarification when it remains necessary. Real-robot experiments show that the framework combines physical information acquisition and userintent clarification within a unified embodied disambiguation process.
Whole-slide visual reasoning requires identifying sparse diagnostic evidence in gigapixel pathology slides and integrating observations across spatial scales. Existing WSI methods either compress densely sampled patches into global representations or use pretrained vision-language models with heuristic region selection, weakening links between predictions and morphology or lacking pathology-trained observation policies. We present AdaptivePath, an active-perception framework that formulates WSI evidence acquisition as sequential decision making. The Navigator learns question-agnostic abnormality-driven navigation from pathologist-reviewed labels to select observation locations and spatial extents, avoiding costly question-specific trajectory annotations. We train this policy through alternating representation learning and proximal policy optimization, followed by fine-tuning with geometric and appearance consistency objectives to stabilize focus trajectories. During inference, the Navigator hierarchically acquires sparse observations from low to high magnification under a limited ROI budget. A Morphology Interpreter converts observations into question-conditioned evidence, while the Deliberator evaluates evidence and revises intermediate answers across magnifications. The Arbiter integrates deliberation history to produce final answers. AdaptivePath achieves state-of-the-art zero-shot performance on WSI and region pathology VQA benchmarks and reaches 80.14% accuracy for cancer subtype classification across six TCGA cohorts. In a blinded diagnostic-utility study, pathologists using AdaptivePath-selected observation sequences achieve 82.9% accuracy. These results demonstrate that learned active perception enables effective and traceable visual reasoning over gigapixel pathology slides.
Spiking point cloud networks usually scan space in a fixed, input-agnostic order, which leaves the most distinctive resource of spiking computation, the temporal evolution of the membrane potential, unused as a locus of decision-making. Active Spiking Perception (ASP) recasts 3D recognition as an iterative decision process in which the network's own leaky integrate-and-fire (LIF) membrane potential, read as a running belief over the class, selects the next chunk to observe and triggers confidence-margin early exit. A lightweight Slice-Selection Policy scores unvisited farthest-point-sampled chunks from the membrane state and precomputed geometric descriptors, trains end-to-end through a straight-through Gumbel-Softmax, reduces to an argmax at inference, and adds about 2% of backbone parameters. We prove that leaky integration is the recursive log-posterior update of a Bayesian filter, that the exit rule attains distribution-free selective risk with no multiple-testing penalty at the stopping time, and that streaming state carry-forward is exactly equivalent to prefix recomputation with bounded finite-precision drift. ASP reaches 90.62% and 93.28% on ModelNet40 and ModelNet10, 1.7 points below the strongest spiking baseline at a larger backbone, while adding a certified anytime interface no baseline offers. The mechanism transfers unchanged to dense prediction, giving 83.21 instance mIoU on ShapeNetPart and 48.50 mIoU on S3DIS Area 5, to our knowledge the first spiking results on S3DIS Area 5, and, fixation replacing chunk selection, to a foveated non-spiking transformer, so the policy is not tied to spiking backbones: cost is exactly linear in observations and the threshold is a measured compute dial spanning 2.8x to 1.35x less energy. One limitation is concrete: one S3DIS class is unidentifiable at the crop size we use, and we give the prediction that would fix it.
Active visual agents solve fine-grained image tasks by interleaving reasoning with image-grounding actions across multiple turns. However, deployment-time rollout budgets are rarely fixed: some requests permit long rollouts, while others require the agent to act under a tight turn limit. Existing methods train the policy as if the rollout budget were hidden, so when the available budget is smaller than the trajectory the agent prefers, the interaction is often truncated before any valid answer is produced; we term this failure \emph{catastrophic truncation}. To overcome this challenge, we present AdaTurn, a budget-aware framework that conditions the agent on the allowed number of turns and explicitly trains the boundary behavior induced by the budget. Our key component, Forced-Answer DAPO (FA-DAPO), converts the over-budget event from a masked or penalized failure into a trainable final-decision step, teaching the model to synthesize partial evidence when further tool use is no longer possible. We further randomize rollout budgets during both training and inference and introduce a load-balanced scheduler that makes such operations practical. AdaTurn substantially improves low-budget accuracy, for example raising VisualProbe-Medium from 36.7% to 47.6% at four turns, while preserving strong scaling at larger budgets and transferring effectively to multiple backbones and general multimodal benchmarks.
Giovanni Pezzulo, Davide Nuzzi, Marco D'Alessandro +3q-bio.NC cs.AI
Recent advances in generative and embodied AI have been driven by large-scale predictive learning over multimodal data. However, the resulting systems remain largely based on passive training regimes where linguistic regularities create the scaffold onto which information from other modalities is attached. Conversely, neuroscience and cognitive science suggest that biological intelligence is organized in the opposite way, where grounded world models acquired through interaction with the environment provide the semantic scaffold to which language is attached. Here, we illustrate five examples of neural circuits supporting grounded world modelling, which underlie navigation in physical and conceptual spaces, affordance-based perception and interaction with objects, active perception and exploratory learning, allostatic control and emotion, and the distinction between self- and world-generated outcomes. These examples highlight several features largely missing from current embodied AI, including the role of intrinsic dynamics as a foundation for learning, the centrality of action in aligning these dynamics with the external world, the prominence of autonomous experience and open-ended learning over passive assimilation of externally provided data, and the fact that early predictive and control mechanisms scaffold higher cognitive abilities such as reasoning, conceptual navigation, planning, imagination, understanding others' minds, and communication. Finally, we discuss whether and how principles derived from biological systems may inform future embodied AI, including training regimes based on social interaction to construct world models that are not only grounded but also socially shared and aligned with human norms and values.
Estimating the full shape of a deformable object is especially challenging when vision is unavailable: in the dark, inside an opaque bag, behind the manipulating hand, or under heavy self-occlusion. Touch is the natural sensor in these settings, but touches are sparse and local. We present a single topology-agnostic estimator that reconstructs the full mesh of a deformable object from only a few touches and no vision, using one permutation-invariant cross-attention architecture that handles a 1D rope, a 2D cloth, and a 3D volumetric soft body. The learned estimator reduces reconstruction error by roughly two-thirds relative to non-learned geometric mesh completion and a Gaussian-process surface baseline, and it outperforms a simpler global-pool set encoder, with the gap growing as more touches are observed. We then show that the estimator's deep-ensemble uncertainty can be used to learn where to touch next, which lowers error further and beats both random touching and a Gaussian-process active baseline at sparse budgets. This gain is modest on average but grows with self-occlusion and on the error tail. When vision is also available, where to touch barely matters, motivating the vision-free setting we study.
Hongkang Yang, Zhi-Qin John Xu, Feiyu Xiong +1cs.AI cs.CL cs.LG
As part of a series on first-principles modeling of cognitive functions, this paper attempts to provide a mathematical formulation of thinking and perception. It formally derives slow thinking or more generally, active perception, and encompasses the design, training and inference of slow thinking large language models. Our starting point is the lifting and projection of probability distributions on the observable and latent spaces, with the objective of representing complex data distributions by simple function families such as neural networks. A theory called "active lifting" is proposed, based on the sampling of latent sequences and an intrinsic drive to reduce uncertainty with maximum rate. It derives a large design space, containing the slow thinking models in a subspace that we call the static theory. These models are positioned on the representation hierarchy and sampler hierarchy induced by the static theory, and can be upgraded by climbing the two hierarchies. Active lifting further derives an inference process with an internal time axis, and a training objective that resembles minimum-length coding as well as the invention of languages. Thus, it characterizes the agency of perception, including the emergence of the slow thinking formats. Technical by-products of this theory include a three-stage pathway for improving slow thinking models, a unified approach to constructing encoders and generative models for all data modalities, a priori formation of human-like visual representations, and a possible solution to policy collapse.
Existing referring segmentation models passively process static images captured from fixed perspectives, limiting their applicability in Embodied AI, where agents must perform active perception in the continuous 360$^\circ$ environments. To bridge this gap, we introduce a novel task: Active Panoramic Referring Segmentation (APRS). In this setting, an agent is required to adjust its viewing direction ($Δθ, Δφ$) to explore the 360$^\circ$ environment, seeking the object specified by a user instruction for segmentation. To tackle this challenging task, we propose PanoSeeker, a memory-augmented agent for efficient APRS. Rather than relying on heuristic scanning, PanoSeeker integrates a Vision-Language Model (VLM) with EgoSphere, an explicit spatial visual memory. By progressively integrating sequential local observations into a unified 360$^\circ$ representation, EgoSphere enables the agent to plan efficient and non-redundant search trajectories. Once the target is found, the agent performs active viewpoint alignment and outputs the segmentation mask. Furthermore, we curate an expert-annotated search trajectory dataset with memory timelines for Supervised Fine-Tuning, followed by Reinforcement Learning post-training to explicitly optimize PanoSeeker's exploration efficiency. Extensive experiments on our newly established APRS benchmark demonstrate that PanoSeeker achieves superior search efficiency and segmentation accuracy, significantly outperforming adapted state-of-the-art baselines.
Boyang Sun, Jiajie Li, Yung-Hsu Yang +6cs.RO cs.CV cs.LG
Autonomous robots often need to move their camera before they can act: to inspect an object, reveal an occluded region, or obtain a view that responds to a user's intent. While vision-language navigation translates instructions to base motion and vision-language-action policies map instructions to manipulation actions, language-conditioned camera motion remains comparatively underexplored as a first-class action. We formulate language-conditioned camera motion generation: given a current RGB observation and a free-form natural-language intent, predict a relative target camera pose for the next observation. This task is inherently non-trivial: viewpoint changes are driven by latent perceptual intentions, and a valid motion may operate at different semantic granularity, from entering a room to looking around a corner, inspecting a visible object, or revealing an occluded detail. To model this structure, we mine multi-intention camera-motion supervision from egocentric video, pairing plausible intents and observation-gain descriptions with relative SE(3) target poses. We propose LIME, a vision-language camera-motion generator that combines an auto-regressive observation-gain output with a continuous flow-matching pose head. This design lets the model jointly predict what the next view should reveal while representing multi-hypothesis target views. Across experiments and downstream robotic tasks, we show that LIME can learn to actively choose camera poses from passive human video, turning ordinary egocentric recordings into supervision for intent-aware active perception.
Certified world models estimate how long their predictions remain valid. We turn this validity horizon into an operational sensing clock: a rule for when an agent should stop coasting and re-sense. Starting from an audited equivariant world model, we derive a deadline for no-sensing intervals and show that deployable deadlines in learned world models must be drift-aware: on-manifold Lyapunov rates alone overestimate coasting validity, while calibrated native rollout-drift envelopes carry the deployed guarantee. On a frozen 3D VN-JEPA model, the resulting clock controls held-out interval-simultaneous certificate violation across seeds and data shards. In a cue-conditioned theorem-bed (a synthetic bench where all schedulers share the exact model, isolating the scheduling rule), the clock remains valid on the deployment distribution and substantially reduces eventful-tail violations relative to exact-mixture expected-belief scheduling at matched sensing budget. We also report limits: in the short-horizon frozen VN-JEPA regime, empirical conformal horizons match the deployed clock on validity and budget, and a partial-reset exploration finds no clean budget-matched advantage for the spectral term. Thus the contribution is a certified sensing-clock primitive and drift-aware deployment method, not a claim that spectral clocks empirically dominate all non-spectral schedulers.
Fatma Youssef Mohammed, Grzegorz Malczyk, Kostas Alexiscs.RO cs.CV
Human visual attention relies on structured scanpaths to efficiently process scenes, yet instilling this behavior into robot autonomy is in its infancy and hindered by the high,computational costs of existing predictive models. To address this, we introduce GazeLNN, a computationally lightweight,scanpath prediction model that leverages Liquid Neural Networks as its recurrent engine and employs MobileNetV3 for feature extraction. Operating auto-regressively, the architecture predicts sequential fixation heatmaps conditioned on the current visual stimulus and fixation history. Despite requiring only 0.61 GFLOPs, GazeLNN achieves state-of-the-art performance on the MIT Low Resolution dataset achieving 0.47 ScanMatch score. It outperforms existing recurrent baselines across diverse evaluation metrics, while reducing computational costs by 99.40% and accelerating inference by up to six times. To investigate the role of human attention modeling in robot autonomy and demonstrate the practical utility of this highly efficient architecture, we integrate GazeLNN into an active camera-robot control policy trained via Reinforcement Learning. This integration enables human-fixation-guided perception during autonomous navigation, validated through successful real-world deployments on an aerial robot.
Zhenghao Xing, Ruiyang Xu, Yuxuan Wang +8cs.CV cs.CL cs.SD
Passive models for long video understanding typically rely on a "watch-it-all" paradigm, processing frames uniformly regardless of query difficulty, causing computational cost to grow with video duration. Although interactive frameworks have emerged, they often rely on global pre-scanning, and their context cost still scales with video length. We propose OmniAgent, the first native omni-modal agent that formulates video understanding as a POMDP-based iterative Observation-Thought-Action cycle. OmniAgent executes on-demand actions to selectively distill audio-visual cues into a persistent textual memory, effectively decoupling reasoning complexity from raw video duration. To operationalize this, we introduce (1) Agentic Supervised Fine-Tuning to bootstrap native active perception via best-of-N trajectory synthesis with dual-stage quality control, and (2) Agentic Reinforcement Learning with TAURA (Turn-aware Adaptive Uncertainty Rescaled Advantage), which leverages turn-level entropy to steer credit assignment toward pivotal discovery turns. Crucially, OmniAgent exhibits positive test-time scaling, where performance improves as the number of reasoning turns increases, validating the efficacy of active perception. Empirical results across ten benchmarks (e.g., VideoMME, LVBench) demonstrate that OmniAgent achieves state-of-the-art performance among open-source models. Notably, on LVBench, our 7B agent outperforms the 10$\times$ larger Qwen2.5-VL-72B (50.5% vs. 47.3%).
Wenhao Lu, Zhengqiu Zhu, Xiaofeng Wang +7cs.CV cs.AI
Aerial Embodied Question Answering (EQA) requires Unmanned Aerial Vehicles (UAVs) to actively perceive the environment and answer natural language questions. Existing outdoor EQA systems usually stop once the target enters the UAV's field of view, leaving the fine-grained viewpoint adjustment needed for evidence-seeking questions largely unresolved. To address this issue, we introduce FG-EQA, a fine-grained active perception EQA benchmark with more than 40K simulated trajectories and 1K real-world trajectories. Drawing inspiration from the ``waggle dance'' of scout bees, which iteratively adjust their flight paths to verify target information, we propose ScoutVLA, an evidence-driven Vision-Language-Action model for outdoor EQA. To emulate this active exploration behavior, ScoutVLA features a decoupled dual-expert architecture: a vision-language expert infers the semantic intent to identify missing evidence, while an independent action expert employs high-DoF flow matching to generate continuous viewpoint-refinement trajectories. To balance the competing demands of continuous control and semantic reasoning, we devise a decoupled training strategy with a knowledge insulation mechanism that prevents the action gradients from erasing the model's multimodal reasoning ability. Extensive simulated experiments and a qualitative real-world field study both verify the superiority of ScoutVLA over the state-of-the-art baselines, demonstrating a 10.48$\boldsymbol{\times}$ higher average strict success rate and a 7.72$\boldsymbol{\times}$ higher average QA correctness.
Michal P. Podolinsky, Neel P. Bhatt, Pranay Samineni +3cs.LG cs.AI cs.MA cs.RO
Perceptual uncertainty is a central challenge for heterogeneous robot teams operating in unstructured outdoor environments, where no single viewpoint affords reliable scene understanding. Perceptual uncertainty, arising from sources such as occlusions, manifests differently across robot viewpoints depending on scene structure. Detecting and resolving sources of perceptual uncertainty requires both scene-based contextual reasoning and capability-aware robot allocation. While vision-language models provide strong semantic priors for both, they are computationally prohibitive for onboard inference and lack calibrated uncertainty quantification. We introduce Co-GLANCE, a real-time onboard perception and decision-making system for uncertainty resolution in heterogeneous robot teams. Co-GLANCE distills the semantic reasoning capabilities of a vision-language model into an end-to-end model for occlusion segmentation and robot allocation, eliminating the need for cloud-based inference. To quantify perceptual uncertainty, Co-GLANCE combines conformal prediction with selective abstention to provide statistically valid coverage guarantees for segmentation, robot allocation, and detection outputs. These calibrated uncertainty estimates directly trigger active perception, dispatching the most appropriate robot to acquire informative viewpoints and resolve uncertainty. Across real-world scenarios, Co-GLANCE outperforms cloud-based vision-language model baselines in occlusion segmentation and robot allocation accuracy by 25% and 36%, respectively, while reducing per-frame inference latency 350x. We also release an air-ground dataset for future research. Code, videos, and dataset available at https://co-glance.github.io/ .
Xingyao Lin, Guojin Zhong, Tianyi Lu +4cs.RO cs.CV
Egocentric human video offers a scalable alternative to robot data for pretraining, yet models pretrained on such video consistently underperform those pretrained on robot data. We attribute this gap to a missing signal, the active perception behavior in egocentric videos, where humans continuously reposition their viewpoint during manipulation, inducing camera motion that standard pipelines treat as noise. To address this, we present ActiveMimic, a pretraining framework that recovers synchronized camera and wrist trajectories from a single body-worn RGB camera, models camera motion as a viewpoint action, and jointly learns active perception and manipulation from in-the-wild egocentric human video before adapting to a target robot. Empirically, real-world experiments across tasks with diverse active perception demands show that ActiveMimic consistently surpasses baselines pretrained on human video and matches state-of-the-art models pretrained on robot data. Further analysis provides evidence that active perception capability originates from egocentric human video pretraining rather than robot-specific fine-tuning, confirming active perception as the key to unlocking egocentric human video for robot pretraining.
Jingsen Zhu, Silvia Sellán, Alexander Terenincs.GR cs.CV cs.LG stat.ML
We develop a framework for task-specific active next-best-view selection in 3D reconstruction from point clouds, by casting the problem in the language of Bayesian decision theory. Our framework works by (a) placing a prior distribution over the space of implicit surfaces, (b) using recently-developed stochastic surface reconstruction methods to calculate the resulting posterior distribution, then (c) using the posterior distribution to carefully reason about which view to scan next. This enables us to perform camera selection in a manner that is directly optimized for the intended use of the reconstructed data - meaning, we reduce uncertainty only in those regions that make a difference in the task at hand, as opposed to prior approaches that reduce it uniformly across space. We evaluate our method across three distinct downstream tasks: semantic classification, segmentation, and PDE-guided physics simulation. Experimental results demonstrate that our framework achieves superior task performance with fewer views compared to commonly used baselines and prior general uncertainty-reduction techniques.