Reliable robotic mapping in dynamic real-world environments is essential for embodied intelligence and long-term autonomy. However, existing methods often involve a trade-off: lightweight online mapping is efficient but captures limited semantic context, while dense mapping provides richer semantics at higher computational cost and is more vulnerable to distractors. They also struggle to relocalize objects after scene changes. To address these challenges, we propose the Online Anchor-centric Scene Inference System (OASIS), a unified system for online mapping, complex open-vocabulary retrieval, and navigation in dynamic environments. Motivated by the observation that indoor scenes are organized around stable semantic anchors, OASIS represents the environment using a sparse set of anchor nodes, each maintaining a unified multimodal representation that integrates semantic descriptions, visual features, and geometric states for reasoning, verification, and navigation. Its compact anchor-centric memory supports on-demand evidence extraction and hierarchical retrieval, reducing redundant storage while improving retrieval reliability. To handle scene dynamics, OASIS further reuses transient observations collected during navigation to relocalize moved objects. Experiments in both simulation and real-world environments, including standard retrieval benchmarks and navigation tasks under both static and dynamic settings, demonstrate that our system can perform complex open-vocabulary retrieval and navigation, while also enabling relocation of displaced objects after scene changes.
https://arxiv.org/abs/2610.30589
Recent advances in foundation-model agents have enabled increasingly capable embodied systems for long-horizon planning and interaction. However, embodied agents require fast, closed-loop decisions as they continuously perceive and act on a changing environment. Deploying strong cloud agents leaves the platform idle for seconds between consecutive actions, while lightweight local models respond quickly but are more vulnerable to errors. Existing hierarchical architectures typically assign models fixed roles, whereas conventional model routers make stateless per-request decisions. We present CARE, a critic-guided adaptive execution framework that dynamically determines when to escalate control from a local executor to a stronger cloud model. A lightweight critic model monitors execution reliability and triggers escalation when local execution becomes unreliable; once escalated, control persists until the agent recovers, avoiding step-wise switching. To support such control transfer, a shared handoff memory maintains execution context and task-level progress across executors, enabling the model to intervene without restarting from scratch. Across diverse embodied benchmarks, CARE improves task success by 6.3-18.3 percentage points over the strongest evaluated collaboration baseline. CARE further approaches or exceeds Cloud-only success while reducing median time-between-action latency by 28.8-82.6%.
https://arxiv.org/abs/2610.30587
Should an embodied agent's experience be baked into its policy weights, or kept external and retrieved at inference? A memory-augmented teacher (a VLM that retrieves lessons, with their episode images, from its own past attempts) produces trajectories on a vision-conditioned reach task; from the same trajectories we distil LoRA student policies and evaluate them under matched protocols, including one that re-attaches the teacher's frozen retrieval buffer to a fixed set of student weights. We pre-registered the hypotheses, arms, tests, and analysis rules before training and before any p-value. Distilling the memory-augmented behaviour into weights beat an action-only control by +1 pp (n.s.); the registered test proves underpowered for its own minimum effect, so we withdraw - rather than refute - the memory-in-weights claim. Attaching retrieval to a fixed set of student weights added +34 pp (z=5.15), reaching the pooled memory-augmented teacher reference (49.3%); the pre-registered protocol contrast agrees (+23 pp, z=3.36); under a pre-registered auto-reset sensitivity count the two are +36 and +29 pp. An exploratory behavioural probe with pre-registered per-arm predictions is consistent with behavioural distillation transferring the marginal correction but not the conditional, situation-specific correction that retrieval supplies. The recipe that follows is to externalise the memory: in follow-up controls registered before they ran, the same frozen buffer lifts the base model with no adapter to 48% (sensitivity 48%) and the other three adapters from 19-26% to 53-55% (sensitivity 17-21% to 50-55%), and the adapter adds nothing detectable once the buffer is attached. On a second, harder task, the retrieval advantage points the same direction but falls short of significance (+6 pp, 95% CI [-4.5, +16.4]); exploratory diagnostics tie the attenuation to memory coverage.
https://arxiv.org/abs/2610.30478
General purpose multimodal language models (LMs) are increasingly capable of directly controlling robotic arms (e.g., by commanding end effector poses) to manipulate their environment. Tracking the development of these capabilities is important for understanding their potential benefits, such as improving industrial automation, and risks, such as enabling malicious users to produce dangerous chemical or biological agents. To this end, we investigate the ability of frontier multimodal LMs to directly control physical robotic arms. We evaluate nine models at multiple reasoning-effort and service-tier settings on six manipulation tasks taken from the RoboDojo benchmark. To our knowledge, this is the first systematic comparison of a broad range of frontier LMs across a common suite of manipulation tasks on physical hardware. We find that current frontier models can directly control real robotic arms to perform a variety of manipulation tasks but are not yet precise or efficient enough to complete them within a practical action budget. However, these capabilities are improving rapidly: performance increases on our benchmark between July and October 2026 (inclusive) correspond to a performance doubling time of 53.9 days, while the best recorded execution rate increased by a factor of 4.4 and the lowest recorded execution cost fell by a factor of 6.2. These trends raise the possibility that the Bitter Lesson is coming to robotics; the capabilities of embodied general purpose LMs may one day be comparable to those of more specialized models.
https://arxiv.org/abs/2610.25287
Multimodal agents (MMAs) sustain interaction through reasoning, memory, tools, and feedback, most visibly as computer-use agents, while robotic systems (RSs) couple sensing and actuation under physical dynamics. We define multimodal embodied agents (MMEAs) as goal-directed systems that couple multimodal task reasoning with physical action and revise decisions from the resulting feedback. This raises a central question: what changes when a multimodal agent moves from computer-use to robot-use? We introduce PAPAV, a unified capability-centric framework of five recurring functions: Perceive the current state, Anticipate action effects, Plan a feasible course, Act through an interface or body, and Verify the outcome. Defined by function rather than architecture, these capabilities let PAPAV identify what all three share and then compare MMEAs with MMAs and with RSs. Five physical constraints, from partial observability to unverifiable outcomes, leave fewer choices fixed in advance and less room to reverse errors. Across 64 benchmarks, explicit evaluation centers on Act; only 5 assess Anticipate and 3 assess Verify, leaving both largely hidden behind task success. These constraints also frame the open challenges we identify. PAPAV therefore provides a common basis for designing reliable physical agents and measuring progress across all five capabilities rather than task success alone.
https://arxiv.org/abs/2610.30108
Tabular studies are a common way to compare independent and centralized multi-agent learners, because they avoid function approximation. A natural expectation is that centralized learning, which learns joint actions, helps agents coordinate, especially when their bodies make solo success hard. Earlier work tests this only in parts: it measures coordination for one learning method, compares methods with the agents' bodies fixed, or varies speed but reports only how well the team did. We test it in a predator-prey environment, where teammates cooperatively hunt together while two teams compete. We compare independent Q-learning (IQL) with central Q-learning (CQL) in all four predator and prey pairings, under limits on speed and stamina, on a 5x5 and an 8x8 grid trained for 200,000 episodes each, and check that the agents learn on both. We measure coordination directly, with predictions for the main measurement fixed in advance. In this task, where one predator can capture prey alone, we find three things. First, CQL costs its agents in both roles on both grids, and its predators coordinate no more than IQL's. Second, its one benefit comes from acting on the best joint action: this cut capture time by 42% on the small grid with faster prey and unlimited stamina, but helped nowhere on the larger grid, where the best joint action had mostly never been tried. Third, on the larger grid a tie-breaking default decided which method won, even though the agents had passed a learning check. A tabular comparison can therefore be trusted only if it also checks how much of the state space the agents covered and implements the methods alike. We propose simple checks for both.
https://arxiv.org/abs/2601.17454
Robots must act under physical and social constraints while adapting to changes in their environment and human needs. Existing embodied-agent systems often focus on specified goals or abstract execution into symbolic actions, limiting evaluation in an evolving world. We introduce TARSphere, a robot-oriented full-stack framework for physically and socially grounded long-horizon embodied agents. TARSphere-Infra runs robot policies in a physics simulator with embodied NPCs and constrained dialogue. An event-driven finite-state machine (FSM) uses completed actions and delivered information to update task states and evaluate outcomes. TARSphere-Harness, its embodied brain harness, lets a frozen multimodal large language model (MLLM) act through multi-level executable skills, using observations, hierarchical memory, and execution feedback to revise its decisions. On this foundation, agent-in-the-loop curation aligns task semantics, causal dependencies, and scene geometry to construct TARSphere-Bench: 360 episodes across 30 indoor scenes. Five paired conditions assess perceptual robustness, disruption adaptation, task update, human coordination, and intent understanding for navigate-and-report and fetch-and-deliver tasks. Our evaluation of 8 MLLMs shows that general-purpose models can outperform embodied models, while task updates and human interaction remain difficult. Frequent collisions and falls further motivate evaluating execution safety alongside task success, even with a shared action interface and decision budget.
https://arxiv.org/abs/2610.30043
Recovering metric world-space hand motion from egocentric video supports understanding human interactions and transferring them to embodied agents. Three challenges complicate this task: multi-stage pipelines propagate errors from independently estimated hand and camera motion; monocular observations leave physical scale ambiguous; and occlusion and out-of-view hands admit multiple plausible motion trajectories. We present DewHand, a unified model for direct joint reconstruction of world-space hand and camera motion. To avoid compounding errors from multi-stage estimation, DewHand jointly predicts hand and camera motion in a shared world coordinate system within a single-stage pipeline. To resolve metric-scale ambiguity, it combines scene-level geometric features from VGGT-$\Omega$ with metric-space supervision, while hand-specific features preserve fine-grained articulation. To handle occlusion and out-of-view hands, we formulate reconstruction as conditional motion generation with flow matching. Under the same architecture, this generative formulation outperforms direct regression in both accuracy and temporal smoothness with only two sampling steps. Experiments on HOT3D and ARCTIC demonstrate improved world-space accuracy and temporal consistency over state-of-the-art methods. DewHand reduces rigid-aligned world-space error relative to StableHand from 32.9 to 28.4 mm on HOT3D and from 76.2 to 36.2 mm on ARCTIC. It also recovers camera scale within 10% on two thirds of HOT3D clips, versus about a quarter for SLAM. Moreover, in zero-shot evaluations on HoloAssist, H2O, and TACO, captured with different devices, DewHand achieves the lowest rigid-aligned world-space error (WA-MPJPE) on each dataset.
https://arxiv.org/abs/2610.29936
Diffusion-based video generation models (VGMs) can produce visually plausible videos, yet their camera ego-motion often drifts over long horizons. Camera ego-motion determines how the viewpoint changes between frames and is essential for maintaining consistent geometry and trajectories. To address this problem, we propose CausalAlign, a causal alignment framework with two components. First, we diagnose how precise ego motion is represented, used, and retained within VGMs using linear probes and causal interventions, revealing a rise-and-fall pattern during denoising. Second, we strengthen ego-motion state in VGM by aligning the representations at the identified layers to a 3D foundation model through delta-token space. This space encodes the changes between consecutive frames, allowing the alignment to focus on ego-motion dynamics rather than static scene content. With Wan2.2 fine-tuned on DL3DV and RealEstate10K, CausalAlign can largely improve motion accuracy, geometric consistency, and temporal smoothness, while preserving visual quality. For example, it improves geometric consistency by more than 20% relative to the base model. Moreover, with the fine-tuned VGM, CausalAlign improves the success rate of world action models (WAMs) by 10 points on mobile manipulation tasks of RoboCasa365 benchmark. Overall, CausalAlign improves ego-motion consistency in diffusion-based VGMs and yields noticeable gains in embodied models.
https://arxiv.org/abs/2610.29915
Multimodal large language models (MLLMs) show promising reasoning capabilities, but reliable physical interaction requires planning that accounts for when actions are executable and how they change the world. These preconditions and effects are environment-dependent and often not captured by MLLMs, making adaptation challenging. Yet existing approaches either require costly environment-specific training or rely on unstructured interaction experience that is difficult to reuse. Therefore, we introduce CausGround, a non-parametric causal interaction grounding framework that distills interaction experience into reusable action-centered causal knowledge. CausGround identifies action preconditions and effects and organizes them in a factorized world-state representation capturing object attributes and inter-object relations. At planning time, task-relevant causal dependencies are grounded in current observations to support backward reasoning over preconditions and forward reasoning over effects for action selection and multi-step planning. We evaluate CausGround on simulated high-level planning, interaction reasoning, and real-world robot manipulation. On adapted VirtualHome and RoboCasa benchmarks, CausGround improves vision-based high-level planning across MLLM backbones, increasing average task success from 24.9% to 53.3% on VirtualHome and closed-loop planning success from 43.0% to 57.0% on RoboCasa. On SWITCH, average interaction reasoning accuracy increases from 25.8% to 31.9%, indicating improved action selection, recovery, and verification. Moreover, real-world experiments show that CausGround reasons backward from MLLM-interpreted goals to derive executable high-level action sequences for sequential manipulation. These results demonstrate that reusable action preconditions and effects support long-horizon embodied planning across diverse environments without parameter updates.
https://arxiv.org/abs/2610.29702
Vision-language-action (VLA) models achieve strong performance across embodied manipulation tasks by unifying visual understanding, language conditioning, and action generation. However, long-horizon tasks can involve non-Markovian dependencies: successful decisions may require retaining task progress and past interaction outcomes, while partial observability and execution errors compound over time. Existing work has begun to address these challenges by augmenting foundation-model policies with memory and by introducing hierarchical architectures that separate high-level planning from low-level execution. We organize the high-level VLM as a robotic agent through an editable harness to maintain task context and improve planning through execution experience. We instantiate this approach as SERA (Self-Evolving Robotic Agent), with a fixed VLA executor providing low-level control. The harness maintains and updates task context using observations and execution feedback to guide subsequent planning. Self-evolution uses environment rollouts to propose and evaluate harness revisions, with the selected configuration fixed before held-out testing. On RoboMME, SERA achieves 51.38% overall success with autonomous selection, compared with 44.38% without self-evolution. Our exploration suggests that agentic planning is a promising direction for history-dependent robotic manipulation.
https://arxiv.org/abs/2610.29560
Safe control is a prerequisite for real-world embodied intelligence, for which safe reinforcement learning has emerged as a promising paradigm. However, existing safe reinforcement learning methods either require costly real-world exploration or depend on hand-crafted safety functions. Neither scales to vision-language-action models deployed in open-world physical environments. We propose SafeDojo, a model-based safe reinforcement learning framework that improves vision-language-action policies through world-model imagination. SafeDojo performs online reinforcement learning on top of an interactive video world model. A ResNet success classifier scores predicted future frames, while a lightweight safety head predicts per-step contact probabilities from the current latent context and proposed action chunk, jointly assessing task execution and trajectory safety. These decoupled signals are balanced through Lagrangian-constrained GRPO. On SafeLIBERO Level I, SafeDojo improves average safe-success rate by 8.25 percentage points over the strongest baseline; on the zero-shot Level II setting, it is competitive with the strongest model-free baseline. Real-world Franka deployment further shows the best average task and safe-success rates across five tasks. These results support world-model-based safe reinforcement learning as a promising path toward safer embodied intelligence.
https://arxiv.org/abs/2606.20698
High-fidelity 4D physical simulation requires accurate physical parameter assignments that align with diverse material properties. Previous methods struggle with critical bottlenecks: they either rely on computationally intensive per-scene optimization or predict parameters from purely visual inputs. The latter suffers from inherent ambiguity, e.g., the inability to distinguish the stiffness of rubber from its appearance alone. To bridge this gap, we propose AutoPhyX, a text-conditioned framework for predicting spatially-varying physical properties. Since training robust feed-forward models requires fine-grained supervision, we first introduce a novel part-controllable data generation pipeline that decomposes complex 3D assets into semantically distinct components and pairs them with physically plausible, diverse parameters and corresponding text descriptions. Leveraging this dataset, AutoPhyX employs a cross-modal modulation mechanism where text features dynamically modulate visual features for precise physical grounding. By formulating property prediction within a voxel field, AutoPhyX ensures compatibility with diverse 3D representations, including meshes, point clouds, Gaussian Splatting, and NeRF. Experiments demonstrate that our method enables physically plausible, text-driven parameter assignment in a single forward pass. It achieves high accuracy and robust generalization to in-the-wild objects, paving the way for future physics-intensive embodied and robotic applications.
https://arxiv.org/abs/2610.29517
Large language models (LLMs) are increasingly integrated into embodied robotic planning, yet standard benchmarks primarily evaluate end-to-end task completion in nominal environments, obscuring failure origins across the execution stack. We present Alpha, a systematic fault-injection and robustness-evaluation framework for LLM-driven manipulation in simulation. Alpha injects calibrated faults across perception (sensor lag), actuation (grip slip), kinematics (unreachable targets), and planning (planner-output corruption), alongside concurrent conditions. To separate physical vulnerabilities from planner sensitivities, we introduce a matched methodology pairing an LLM agent (Llama-3 via Ollama) against a deterministic Scripted Oracle planner across a balanced 150-episode matrix under identical seeds, geometries, and safety criteria. Across the matrix, the Oracle achieves 64.0% (48/75) success while the LLM achieves 44.0% (33/75) (paired McNemar p = 3.01 x 10^-4). Crucially, excluding planner-output corruption, performance across all physical and kinematic conditions is identical: 55.0% (33/60) for both planners with zero discordant pairs (McNemar p = 1.00). This demonstrates that physical actuator slip, kinematic reach boundaries, and perceptual staleness affect deterministic controllers and LLM planners equivalently. The performance gap is driven entirely by planner-output corruption (100% Oracle vs. 0% LLM), isolating the prompt-to-JSON serialization and validation interface rather than a cognitive reasoning deficit. We also audit 169 historical pilot attempts for transparency.
https://arxiv.org/abs/2610.30197
Thermal cameras enable robots to perceive human activities in darkness while revealing less personally identifiable appearance information and fewer details of the surrounding environment than RGB cameras. As vision-language models (VLMs), originally developed for RGB imagery, are increasingly integrated into embodied systems for scene understanding and high-level planning, thermal sensing offers a promising way to extend activity perception to low-light and privacy-sensitive settings. However, the failure modes of VLMs in thermal human activity understanding for human-robot interaction (HRI) remain underexplored. We fill this gap by evaluating seven locally deployable VLMs, including models developed for physical AI and robotics, on two person-centric thermal datasets. Our diagnostic tasks examine fine-grained action recognition, activity timeline reconstruction, and the ability to withhold unsupported answers. Paired RGB recordings serve as a performance reference, while controlled variations in frame sampling, thermal input presentation, and reasoning are used to assess whether the observed errors can be mitigated. Our results show that models can recognize broad activity categories but struggle with fine-grained actions such as coughing, sneezing, and yawning. They also produce inaccurate event timelines and hallucinate actions involving objects. Simple thermal input preprocessing and additional reasoning offer only limited, model-dependent gains. Abstention behavior also varies across models even under explicit instructions, limiting their reliability for practical robotic use.
https://arxiv.org/abs/2610.29244
World models predict how embodied systems evolve under actions, but their accuracy can degrade when physical conditions move outside the training distribution. We study two components of causal modeling: which inputs a predictor may use, and how its mechanism computes their effects. In a synthetic benchmark with shortcut features, a frozen dependency mask derived from paired interventions reduces mean out-of-distribution (OOD) RMSE from 1.769 for a fully connected predictor to 0.697, close to the generator-oracle mask (0.694). For robot link-mass extrapolation in CausalWorld, with disjoint training and test mass ranges, we develop finger-local world models in which link masses enter inertia and dynamics-bias terms through explicit affine coefficients, actions act through either a learned torque network or the known PD controller, and learned residuals capture contact-related effects. Compared with a separately trained control whose mass inputs are fixed at training means, the known-PD model turns out-of-range masses into a consistent predictive gain: median time-averaged joint R2 over 150-step rollouts rises from 0.609 to 0.727, with improvement in all ten task-seed runs (exact two-sided sign test, p = 0.002). In a shared 100-step forecast, the learned-torque model is competitive with the strongest recurrent robotic world model (RWM) configuration (median joint R2 of 0.755 - 0.768 versus 0.752), whereas supplying out-of-range masses sharply degrades the MLP and recurrent predictors. These results indicate that how a physical parameter enters the dynamics mechanism determines whether its out-of-range value helps or harms prediction.
https://arxiv.org/abs/2610.29088
Coding agents are rapidly advancing from single-file generation to long-horizon repository-scale engineering, yet the evaluation of their capabilities to author interactive extended reality (XR) applications remains an open challenge. Unlike traditional software, XR applications operate within dynamic 3D physical environments where runtime robustness depends jointly on continuous user motion, spatial input, and environmental sensing. We introduce GenXR-Bench, the first benchmark for the end-to-end evaluation of coding agents in XR software development. GenXR-Bench contains 112 expert-authored tasks spanning 24 feature implementations, 21 bug repairs, and 67 application generation tasks. Held-out tests combine scripted and adaptive interactions with programmatic and visual assertions to evaluate core functionality and runtime behavior. GenXR-Bench's testing suite achieves 96.5% accuracy against blinded expert judgments of 124 candidate applications executed on consumer XR hardware. Evaluating 10 models on GenXR-Bench, we find that the strongest configuration achieves only 50.9% overall Pass@1. Failure analysis identifies recurring misunderstandings in asynchronous sensor lifecycles, physical grounding, and embodied interaction semantics. GenXR-Bench establishes a reproducible testbed for evaluating coding agents in the interactive and physically situated conditions of spatial computing.
https://arxiv.org/abs/2610.29014
The explore/exploit trade-off is the core problem of reinforcement learning (RL). We show that a minimal foraging swarm approaches its optimum - the ideal-free / Nash allocation of a multi-agent resource-sharing game (an equilibrium, not a return maximum) - with no learner and no reward signal. Agents explore by heavy-tailed (Levy) flights, recruit neighbours to a discovery through a threshold cascade on their proximity network, and disperse as a patch crowds; this cascade cycle balances exploration and exploitation and drives foragers toward the ideal free distribution. The swarm attains the shape of this optimum (committed foragers track richness with slope 0.94) but not its exact equilibrium - it stays measurably exploitable - reaching it dynamically (~150 steps) through the coupling of recruitment to movement: remove the coupling and matching collapses (0.94 -> 0.08). The RL sits at the level of the problem and the allocation reached, not the micro-rule: no step evaluates payoff (reward-free by construction, though the crowd signal is informationally tied to payoff), and the swarm reaches the allocation without learning it (quality plateaus and re-settles after resources move, no cumulative gain). Against a trained baseline the comparison is two-sided: in an easier, disembodied game a learner is more precise, while in the same embodied environment the reward-free swarm reaches ~65% of a validated oracle ceiling at zero training and leads until trained PPO and centralised-critic (CTDE) learners overtake it (~74-75%, 16 seeds, all above the swarm mean) after ~4,000 of 15,000 training steps - its advantage is in budget, not asymptote. In this one setting we read it not as beating RL but as a strong zero-shot, reward-free operating point: a demonstration that the self-organizing physics of a minimal swarm can carry a collective close to the optimum RL seeks, with none of RL's machinery.
https://arxiv.org/abs/2610.25113
Motivated by the human ability to integrate experience across language, perception, and action, we introduce UniWALM, a unified world-action-language model that learns video, action, and language in a single shared representation, where each modality is both a conditioning signal and a supervision target. We ask: can jointly generating language enrich this shared representation, improving manipulation and supporting memory-dependent behavior? Unlike prior approaches that attach a reasoning or vision-language model to a video-action backbone and run them as distinct stages, UniWALM unifies all three modalities inside a single mixture-of-transformers backbone whose per-modality diffusion heads co-denoise synchronously, decoding subtask and textual memory in the same pass as action with no separate planning stage; the generated language is re-consumed on the next step, closing a language-as-memory loop without an external planner or memory bank. To keep the action stream reliable under this richer objective, we further introduce Flow Matching with Correction (FMC), a training objective that teaches recovery from perturbed action trajectories at no inference-time cost. UniWALM consistently improves over same-backbone baselines (e.g., FastWAM) on compositional pick-and-place (MolmoSpaces) and across the memory-dependent subset of RMBench, where it reaches 34.5% without any pretrained language or reasoning model and exceeds Mem-0's keyframe-equipped setup (28.5%) using only a single anchor frame. On a real-world Find-the-Item task, generating textual memory lifts success from 11.1% to 61.1%, while state-of-the-art VLAs stay at or below chance (<=33.3%).
https://arxiv.org/abs/2610.28626
Object-goal navigation (ObjectNav) requires an embodied agent to find an instance of a specified category in an unseen environment and stop near it. The agent typically relies on an object detector but must judge online from a limited egocentric video stream, sometimes a single frame, whether a detected candidate is the target. Detectors often confidently mistake similar objects, such as a sofa for a bed, and stopping at a false positive fails the episode. Prevailing systems also approach candidates greedily, leaving them occluded or partially observed. We present CLOSER, an evidence-aware active verification framework that treats stopping as closed-loop evidence acquisition, coupling vision-language model (VLM) verification with active perception. CLOSER fine-tunes an open-source VLM to label each view of a candidate tracked across frames as target, non-target, or uncertain, selects complementary views with an embedding-based diversity gate, and fuses their per-view probabilities with a lightweight network. If the fused judgment remains uncertain, a diffusion policy drives the agent toward the candidate using only the tracked mask and depth until the decision becomes decisive. CLOSER runs locally and plugs into existing ObjectNav systems without retraining them. On Gibson and Habitat-Matterport 3D, it raises average success rate by 3.6 points and success weighted by path length by 2.0 points, showing that verifying evidence before committing improves embodied navigation.
https://arxiv.org/abs/2610.28476