autonomous vehicle teams

Abstract: Occupancy predictors rely on large persistent banks of Gaussians or learned queries whose size is fixed by hand. Common pruning proxies, such as opacity, activation frequency, or parameter magnitude, do not measure whether removing a row changes the model's voxel decisions. We introduce Decision-FLIP, a budget-free compilation framework whose scoring and structure selection need only unlabeled calibration inputs. Exact Mean-FLIP scores each persistent row by the number of voxel argmax decisions that change when the row is removed from the readout; a guarded Otsu split of the log-score distribution selects the survivor count without a target and abstains when the split is unsupported; and the survivors are physically compiled into a smaller checkpoint, so camera sampling, attention, and aggregation shrink with the bank. For streaming banks, scores are integrated over each row's lifecycle through warp, death, and refill. Across four Gaussian-bank model-dataset settings and one persistent-query transfer, the automatically selected cuts remove 12-79% of eligible rows and cut bank/query-dependent latency by 7-45% across settings with component profiles, with whole-model savings of up to 8.0% on the primary endpoints bounded by fixed-cost image, voxel, and sensor modules and peak-memory savings of up to 16.6%. Under matched one-epoch recovery controls, semantic mIoU changes by at most 0.18 point. Same-$K$ controls show that the ranking, not only the bank size, matters: random removal collapses GaussianFormer-1 and OccFormer, while label-free Mean-FLIP matches or outperforms supervised Taylor ranking on all five settings where that control was run. The method applies when removable primitives keep a stable checkpoint-row identity; it is budget-free in survivor count, not parameter-free.

Abstract: A multimodal agent must decide whether visual evidence is worth acquiring, which candidate or resolution to query, and when further interaction can no longer change its decision. We study this problem under a strict free-metadata/paid-observation boundary, stochastic vector costs, and a frozen execution trace, and treat it as decision-certifiable active evidence acquisition. A Counterfactual Evidence Interaction Graph (CEIG) represents unary evidence value, signed second-order interaction, and directed conditional harm while maintaining explicit score enclosures for unresolved interactions; a decision-weighted enclosure objective tightens the bounds that can change action identity. Gap-directed refinement evaluates only interactions that can displace the incumbent acquisition and stops once an $ε$-decision certificate shows that unresolved terms cannot change the selected action beyond a declared tolerance. The certificate is model-relative: it verifies policy-consistent optimality under the pinned learned score, and semantic quality is measured separately through post-freeze labels. Paired continuation rollouts learn the downstream value of paid previews, an online operational cost filter is separated from independent finite-policy-family trajectory certification, and an acquisition monitor with an independently implemented verifier binds every certificate to the frozen trace. On 640 cases split into 384/128/128 training, validation, and test cases, the controller reaches task score 0.842, grounding F1 0.816, evidence recall 0.873, 1.74 accepted calls, 286.5K visual tokens, and 742ms p95 latency. Dense replay of issued certificates shows 0.992 action agreement and a 0.988 certificate rate while evaluating 19.6 instead of 191.4 expensive interaction terms per decision.

Abstract: To reach top accuracy, current fully supervised instance segmentation methods severely rely on large-scale pixel-wise labeled datasets. They are usually expensive and time-consuming to obtain. Though weakly or semi-supervised methods utilize cheap bounding box labeled, image-level labeled or unlabeled samples to save the labeling cost, their performance is largely sacrificed. To save labeling cost without losing much performance, in this paper, we present a pipeline that can utilize economical bounding box labels and accurate pixel-wise labels in a hybrid way. Specifically, we design two ancillary models to learn label noise suppression and obtain accurate pseudo pixel-wise labels from bounding box labels for training. One is designed to suppress mislabeling between foreground and background, and the other is designed to suppress noise from mislabeling of instances. Moreover, we exploit category-aware spatial attention module, category constraint module, instance constraint module, and self-learning training approach to improve the accuracy of pseudo labels. Experiments on the PASCAL VOC 2012 and the Cityscapes datasets show that our method can achieve competitive performance with much less labeling cost.

Abstract: Cooperative 3D tracking must integrate complementary observations across agents and time while maintaining consistent identities. When evidence integration and identity inheritance share a matching decision, errors arising from cross-view appearance differences and spatial misalignment can compromise both feature fusion and track continuity. We propose FFBL-Coop, a fuse first, bind later framework that separates instance admission from identity management. Confidence-ranked Slot Admission (CSA) allocates cooperative queries to available ego slots using confidence and spatial proximity. Unified Representation Aggregation (URA) uses cooperative semantic features and aligned anchors to guide ego-feature retrieval, refining the augmented query bank within a shared transformer decoder. After refinement, Cooperative-Priority Identity Anchoring (CPIA) combines learned association with persistent mappings to establish accepted identity assignments across frames. A shared codebook reduces transmitted payload while retaining AP and AMOTA close to the uncompressed variant. FFBL-Coop achieves AMOTA/AP of 0.611/0.548 on V2X-Seq and 0.688/0.653 on Griffin-25M. Code will be released.

Abstract: Self-supervised learning draws inspiration from infant visual development, yet standard training pipelines bear little resemblance to it: images are independently sampled and globally shuffled across epochs. We study self-supervised learning from continuous video streams, where frames are consumed in temporal order using strict sliding-window batches, without global reshuffling or multi-epoch replay. To this end, we construct WT++, a 95-hour urban walking-tour video dataset for streaming pretraining. Combined with a comprehensive evaluation suite we find that contrastive and distillation-based methods struggle in this setting, while MAE is more robust but still falls short of standard i.i.d. pretraining. We find that high inter-batch similarity, caused by sliding-window consumption across consecutive batches, does not explain this gap. The main challenge is high intra-batch similarity, where frames within each batch are near-duplicates. To mitigate this, we propose StreamMAE, which preserves the core MAE reconstruction objective while adapting the input pipeline with stream-aware regularization and motion-biased crop selection. StreamMAE outperforms streaming baselines, matches i.i.d. MAE trained on the same video data, remains competitive with ImageNet-pretrained MAE, and scales positively as the pretraining stream grows from 12 to 95 hours.

Abstract: Embodied systems must make knowledge acquired during one encounter usable in another despite changes in viewpoint, motion, and illumination. Yet aggregate cross-video accuracy conflates failures of local perception with failures to preserve observation identity, establish correspondence, and compose evidence, obscuring whether local video understanding actually transfers. We introduce EgoGears, a complementary single- and multi-video benchmark designed to diagnose this transition. It contains 567 single-video and 1,487 multi-video questions derived from 126 human-collected egocentric recordings covering 39 outdoor routes. Repeated traversals across movement speeds and lighting conditions ground comparisons in shared physical environments; 531 questions require alignment across independent recordings. Single-video questions measure the local visual, spatial, and motion evidence available to a model, while multi-video questions test whether evidence remains bound to the correct observation and can be composed into consistent route relationships. We report 29 single-video and 31 multi-video MLLM configurations across six model families in the main leaderboard. Among the 20 configurations evaluated comparably on both splits, every model performs worse on multi-video questions, with a mean decrease of 22.5 percentage points, and the gap persists when answer format and scoring are held fixed. The gap is not explained simply by additional videos or recording boundaries. The central bottlenecks are observation--evidence binding and ordered route-state tracking. The code and benchmark are publicly available at https://github.com/lei-qi-233/EgoGears.

Abstract: Latent world models often struggle with long-horizon planning despite accurate short-term predictions. Recursive rollouts accumulate errors, while distance concentration in high-dimensional latent spaces can weaken goal discrimination. We introduce the Dual-Latent World Model (Dual-WM), which separates local execution and long-range planning through distinct state representations and dynamics models. The low-level model predicts action-conditioned transitions, while the high-level model uses learned macro-actions to plan over longer temporal spans. We also propose Long-Horizon Representation Learning with Weighted Rollout (LoRe), which supervises self-generated predictions at both levels. An analysis of recursive error propagation motivates exponential horizon weights with separate decay rates for the two temporal scales. During planning, the high-level model generates latent subgoals that the low-level model refines into actions for precise execution. We evaluate from-scratch Dual-WM on five goal-conditioned visual control tasks against the task-wise strongest baselines without actor-guided proposals. At goal offsets of 50 and 100 environment steps, mean success increases from 75.9% to 84.4% and from 61.4% to 69.5%, respectively. At offset 100, Dual-WM outperforms these baselines on all five tasks and improves mean success over LeWM by 30.8 percentage points. Ablations and supporting analyses provide evidence of more informative representations for goal evaluation and greater consistency under recursive prediction. These results highlight the value of separating temporal roles and training across multiple horizons for reliable latent planning. Our core implementation is available at https://github.com/DeLin1001/Dual-WM-Official.

Abstract: Emerging optical-network applications increasingly use received data for inference and control rather than exact source reproduction, creating an opportunity to trade bit-level fidelity for greater transmission reach and efficiency. We propose an end-to-end optical semantic communication system for joint image classification and reconstruction over a nonlinear wavelength-division multiplexed (WDM) fiber channel. The system maps each image directly into a fixed-length sequence of channel symbols that preserves task-relevant information, without explicit source compression or channel coding. Experiments on the MNIST dataset cover launch powers from -9 to +3 dBm, fiber lengths up to 800 km, and 16-, 64-, and 256-Quadrature Amplitude Modulation (QAM) formats. At 0 dBm, classification accuracy remains between 98.92% and 99.31% across all tested link lengths and modulation orders, while requiring fewer transmitted symbols than a Low-Density Parity-Check (LDPC)-coded JPEG baseline at every tested modulation order. These results show that semantic communication can simultaneously extend optical reach and reduce transmission resources by conveying only task-relevant information.

Abstract: Deep vision models often degrade under distribution shift. Test-time adaptation can improve robustness but typically requires iterative optimization, hyperparameter tuning, and multiple forward-backward passes. We propose Zero-Training Fisher Geometry Alignment (ZFGA), a closed-form method that improves robustness under covariate shift without modifying model parameters. ZFGA is based on the observation that distribution shifts distort feature-space geometry. It estimates the Fisher information matrix of the predictive distribution with respect to feature embeddings and applies a linear transformation that aligns test-feature Fisher geometry with a reference geometry computed from clean data. This provides a natural-gradient-inspired preconditioning step in feature space. We evaluate ZFGA on CIFAR-10-C and ImageNet-C using ResNet-50, DINO ViT-S/16, and CLIP ViT-B/32. ZFGA consistently improves over zero-shot inference across all three models, although it is not the strongest method for every model. Covariance whitening performs better on ResNet-50, while Fisher whitening is statistically indistinguishable from ZFGA on CLIP. Across six training-free and gradient-based alternatives (covariance whitening, Fisher whitening, TENT, T3A, LAME, and AdaNPC), ZFGA is the only method that does not substantially harm any of the three model families. The Fisher geometry distortion is also positively correlated with ZFGA gain (Pearson r = 0.366, p = 0.017), providing preliminary evidence that geometric misalignment contributes to robustness degradation. ZFGA requires only forward passes and matrix operations at inference time, offering a lightweight and deterministic alternative to optimization-based test-time adaptation.

Abstract: Diffusion and flow policies can model complex behaviors in offline reinforcement learning (RL). However, penalizing their KL divergence from the behavior policy can discourage actions having high critic values with low behavior density. Directly refining behavior proposals may be an alternative, yet Gaussian or deterministic editors limit expressiveness to represent multiple separated modes for the same proposal. In this work, we introduce Proposal-Conditioned Refinement Flows (PReFlow), a policy extraction method combining critic-based proposal selection with a conditional refinement flow. To optimize proposal selection and refinement together, we formulate a KL-regularized objective whose optimum induces a Gibbs policy over final actions under a Gaussian-smoothed behavior prior. The refinement flow can represent multiple high value modes, while a proposal-centered Gaussian reference regulates large action changes. This Gaussian reference further enables us to make use of simulation-free, closed form adjoint matching targets from sampled endpoints and critic gradients, yielding a single velocity regression loss without a backward adjoint solve. On 50 OGBench tasks, PReFlow achieves competitive offline performance and the highest aggregate score among the compared methods after online fine-tuning, reaching 91\% after 500K environment steps.

Abstract: Language-conditioned trajectory generation is here, but its evaluation has not kept pace. Existing pedestrian trajectory metrics compare trajectories with real-world human data. This does not scale to text-to-trajectory generation across diverse contexts, as collecting human trajectories for every scenario is costly and infeasible. Moreover, pedestrian behavior is heterogeneous and context-dependent, with no single metric as the correct answer, and current evaluation frameworks are not transferable to this domain. These challenges make scalable, reliable evaluation difficult. We introduce STRIDE, the first framework for evaluating context alignment between scenario descriptions and pedestrian trajectories. STRIDE addresses these challenges through three design choices. First, we derive our VRDST evaluation protocol from sociological theories to define a complete evaluation space. Second, it decomposes high-level context into scenario-adaptive behavioral questions. Third, every question is resolved against a deterministic measurement tool library that yields reproducible answers. Together, STRIDE enables complete, verifiable, automated, and scalable evaluation across diverse contexts without requiring human trajectory data. We instantiate STRIDE in the crowd domain as STRIDE-Bench, comprising 1K scenarios, 6K behavioral questions, and 11K measurements with calibrated expected answers across 30 real-world maps. Comprehensive human validations show that STRIDE-Bench is consistent with human behavior and judgment, achieving 80% human agreement. We further evaluate several text-to-trajectory models, finding limited context-alignment capability and persistent challenges in fine-grained context conditioning. We believe that the STRIDE framework provides a first step toward principled evaluation of context-aligned pedestrian trajectory generation.

Abstract: Chain-of-Thought (CoT) reasoning is increasingly incorporated into Vision-Language-Action (VLA) models, yet it can degrade the performance of stronger embodied agents. We investigate this capability-dependent effect by distinguishing explicit reasoning from latent decision computation, i.e., perception-grounded computation that directly supports action prediction. Under the standard \textit{think-then-act} (TTA) paradigm, intervening on the generated CoT while fixing the visual input and model parameters causes a substantial performance collapse, with Navigation F1 dropping from 72.14% to 11.84%, demonstrating the strong influence of explicit reasoning on action generation. We then propose \textit{decide-then-explain} (DTE), which predicts actions before generating explanations, and introduce Visual Conditional Contribution (VCC) and Reasoning Conditional Contribution (RCC) to characterize the resulting decision process. Across autonomous driving and robotic manipulation, DTE consistently outperforms TTA and conventional \textit{no-CoT} baselines, while exhibiting greater reliance on perception-grounded computation. Further TTA-trained, DTE-inference experiments show that this benefit is not solely attributable to retraining under the new factorization. Our results suggest that for capable embodied agents, explicit CoT may be better used to shape decision computation during training rather than mediate action generation at inference time. Code: https://github.com/ocean-luna/openvla-decide-then-explain.

Abstract: Off-policy evaluation is a foundational component of offline reinforcement learning, aiming to assess and optimize policy performance using pre-collected datasets. However, such datasets often suffer from pronounced challenges, including distribution shift, $Q$-value overestimation, and low sample utilization efficiency. To address these issues, this paper introduces a weighted Bellman residual minimization framework that incorporates density ratio weighting by effectively integrating expert demonstrations with behavioral data. The proposed weighting scheme departs from the conventional completeness assumption commonly imposed in the theoretical analysis of deep reinforcement learning. We establish a sharp convergence rate for density ratio estimation and derive the convergence rate for the excess risk of resulting deep $Q^*$ estimator. Extensive empirical evaluations demonstrate that, compared to existing methods, our method achieves significant improvements in numerical performance and policy generalization, providing specific guidance for the rational utilization of expert demonstrations.

Abstract: Weakly Supervised Semantic Segmentation (WSSS) learns pixel-level predictions from image-level tags. Recent work focuses on improving coarse CAMs extracted from large vision-language models (commonly CLIP), but does little to improve their accuracy along segment boundaries. That job is instead delegated to a post-processing method like DenseCRF. However, because DenseCRF relies on low-level colour cues, it can flip correct labels to incorrect ones when neighbouring pixels share similar colours. SAM has recently been adopted as a natural alternative, yet it simply takes on DenseCRF's role as an intermediate "refinement" step that outputs one-hot pseudo-labels in prior work. By discarding the valuable uncertainty in CAMs, these one-hot pseudo-labels turn borderline errors into confidently wrong targets. Our key insight is that CAMs should supervise training alongside SAM boundaries, each through its own loss, rather than being fused together into a single hard target. Inspired by CRF potentials, we propose a framework that disentangles soft pseudo-labels as unary supervision and binary edge maps as pairwise supervision. We realize our framework in a single-stage model, DS-CRF, using CAMs from dino.txt and boundaries from SAM. DS-CRF sets a new state-of-the-art of 56.5% mIoU on MS COCO.

Abstract: Point cloud autoencoders are fundamental components for 3D world representation and support many safety-critical downstream applications. Existing studies have extensively investigated backdoor attacks on point cloud classification, whereas backdoor attacks against point cloud autoencoders remain largely unexplored. However, their backdoor behaviors differ substantially due to the intrinsic structural gap between discriminative and generative models. Specifically, a classifier is a discriminative model that separately fits the marginal distributions of benign and malicious data. In contrast, the generative nature of an autoencoder entangles the two within a unified latent distribution, leading to information crosstalk and reduced attack controllability. In this setting, residual source geometric information in malicious data may leak into the clean inference branch, causing the reconstruction to collapse toward the source data. We then propose the Information Blackhole principle, which introduces Gaussian distribution constraints to disentangle latent representations and block interfering information. Building on this principle, we further propose Adaptive Gaussian Matching (AGM), which explicitly regularizes the latent distribution of poisoned samples. By suppressing the propagation of source geometric information from poisoned features to the attacker-specified reconstruction target, AGM improves attack controllability. Extensive quantitative and qualitative experiments on ModelNet and ShapeNetPart demonstrate that the proposed framework improves the attack performance of several standard triggers and reveals the unique operating mechanisms of backdoor attacks against point cloud autoencoders.

Abstract: In-context reinforcement learning (ICRL) enables large language model agents to adapt to new environments using their interaction history without updating model parameters. However, repeatedly conditioning on growing histories can lead to substantial token cost. We propose a bounded-history context-management framework that predicts the task-dependent downstream effect of removing historical interactions to guide history selection and determine a decision-dependent context budget. Formally, our framework uses the full rolling history as a reference. The predictor evaluates removal effects, defines a deletion ordering, and applies a shared selection criterion to determine how much history to retain at each decision. We evaluate the method in closed-loop SUMO driving under held-out in-distribution, unseen-domain, and unseen-route settings, and in ScienceWorld under a continual ICRL protocol. Relative to a baseline using the full context, our method reduces total token usage by 25.7%, 25.8%, and 23.2% across the three driving settings while maintaining comparable closed-loop driving performance. In ScienceWorld, it reduces total token usage by 52.1% compared to full context and uses 30.2% and 37.8% fewer tokens than the Recent and Similarity baselines, respectively, while maintaining performance.

Abstract: Reinforcement learning (RL) trains representations on data selected by the agent's policy, which then uses the resulting returns to guide its next choices. We show that this loop can sustain a lower-return policy even when representation fitting is globally optimal on those data. In a self-confirming superposition trap, every optimal code assigns overlapping directions to features that rarely occur together under the current policy. An alternative action brings them together, causing interference that lowers its return and reinforces avoidance, although refitting to that action would yield more return at the same capacity. We characterize the dimensions admitting a trap in a tied two-step model and show separately that equal feature frequencies, continued visitation, and independent controller learning need not prevent it. Because fitting weights errors by visitation, an avoided action can lose its return advantage at little cost to the objective. In a finite-action model, we bound this distortion and derive a replay condition: sufficient training weight on the best separately adapted action preserves its ranking despite residual error. Neural PPO experiments show how the feedback develops during learning: agents initialized toward different actions develop different interference patterns, opposite mean return rankings, and different final policies at the same capacity. We therefore test whether retaining access to neglected states can improve control. Keeping these states in training reduces measured interference and improves sequential return, with gains even when the encoder is frozen. Related interventions on state access, replay weights, and feature overlap improve control on MiniGrid and DMControl. For agents that learn through a world model, protected fitting improves DreamerV3--Crafter's cumulative training scores at unchanged capacity.

Abstract: We propose a self-supervised framework designed to enhance the capability of scene text detectors in identifying and recognizing text in scenarios where instances are shown at significant distances, typically small, blurred, and frequently missed by conventional models. Our approach leverages the high-fidelity performance of existing text spotting models on large, clear text as a foundational supervisor. By temporally back-tracking these high-confidence detections through video sequences, we automatically synthesize pseudo-labels for preceding frames where the distant text is still visually degraded or undersized. These pseudo-labels enable training a student model specialized for early text detection, without requiring any manual annotation. The success of this approach depends on accurate pseudo-label generation, for which we develop a dedicated scene text tracker capable of maintaining consistent text identities across challenging video sequences. In addition, we propose SceneText50, a diverse multilingual outdoor dataset to facilitate training and evaluation. Experiments show that our framework significantly improves early detection accuracy and robustness across varied scenes and languages. Code and data are at \href{https://github.com/trid2912/BackTrackingText}{https://github.com/trid2912/BackTrackingText}.

Abstract: Multimodal test-time adaptation (TTA) aims to adapt a pretrained multimodal model online to distribution shift across modalities using unlabeled test data, showing broad potential in real-world applications. However, existing methods primarily focus on adjusting fused features to bridge the source-target gap, lacking explicit control over intermediate representation misalignment, which is a key driver of performance drop under distribution shift. In this work, we tackle this challenge from the perspective of representation engineering. Unlike previous TTA methods that update fusion weights in place, we propose FourIer Representation Editor (FIRE), a novel multimodal TTA approach that directly edits semantically rich intermediate representations. Specifically, we first adopt representation editors into each intermediate layer of the unimodal encoders, enabling layer-wise calibration of unimodal representations. To further enhance the diversity and stability of the low-rank editing subspaces, each representation editor performs frequency domain mixing via the fast Fourier transform to construct structured bases. Moreover, we introduce multi-level adaptation objectives to optimize these editors, jointly promoting cross-modal semantic alignment, source-target statistical alignment, and asymmetric prediction consistency. In this way, FIRE yields aligned unimodal representations for fusion and further improves prediction reliability. Extensive experiments on two widely used multimodal benchmarks under various corruption types demonstrate the superiority of FIRE over existing multimodal TTA methods.

Abstract: Constrained general-sum dynamic games are a popular formulation for highly interactive multi-agent planning problems. In recent years, Generalized Nash Equilibrium (GNE) solvers have achieved real-time performance for small dynamic games. However, solution speed still remains a bottleneck, and controlling even a small number of agents (e.g., more than four) in a dynamic task remains elusive. In addition, Newton solvers that focus on the first-order conditions are vulnerable to non-Nash saddle points, limiting the usefulness of the provided solution. In this work, we propose a fast and versatile interior point solver for constrained dynamic games, along with a computationally efficient second-order correction that increases the probability of converging to a local GNE solution in constrained dynamic games. The performance of the proposed method is evaluated on numerical benchmarks and a physical experiment involving scaled race cars.

Abstract: Polarization imaging captures distinctive surface and geometric cues that benefit a wide range of vision tasks. However, real-world polarization acquisition is often affected by multiple coupled degradations, making image restoration essential for practical polarization vision. Existing methods are largely tailored to specific degradations and remain constrained by the limited scale and quality of polarization data. To address these limitations, we develop an all-in-one polarization restoration framework for diverse and composite degradations. We first study the impact of different polarization representations on restoration performance and identify the normalized Stokes representation as an effective choice for separating intensity and polarization information. Accordingly, we devise a dual-branch architecture that separates intensity and polarization modeling. To overcome the limitations of polarization-specific training, the intensity branch leverages pretrained general restoration priors and a mixture-of-experts extension for composite degradations, while its restoration knowledge is adaptively distilled into the symmetric polarization branch via a cross-domain feature transform. In addition, we establish a composite-degradation polarization benchmark to support all-in-one restoration research. Extensive experiments on public datasets and our proposed benchmark demonstrate the effectiveness of the proposed method.

Abstract: Large language models (LLMs) have recently demonstrated promising capabilities as in-context policy optimizers for Reinforcement Learning (RL), enabling policy search driven by both numerical reward signals and natural language reasoning. However, deploying such methods in practice requires transmitting raw policy parameters and rewards history to cloud-based LLM APIs, exposing proprietary control strategies to third-party service providers. To address this issue, this paper introduces Privacy-Preserving Prompted Policy Search (PP-ProPS), a framework that enables LLM-guided policy optimization while keeping policy and environmental parameters confidential. PP-ProPS encodes policy parameters and reward values using secret client-side transformations before they are included in each API request, ensuring that the LLM provider observes only encoded policy parameters and scaled reward information. Furthermore, unlike Vanilla ProPS, the proposed framework does not require the true optimal episodic return to be known or disclosed to the LLM. Beyond protecting the optimization data, PP-ProPS improves the search process in two ways. First, it provides the LLM with individual reward components instead of only a single total return, offering more informative feedback about each candidate policy. Second, it uses a bounded history that prevents the prompt from growing indefinitely, improving search with high-dimensional policies and supporting the use of open-weight LLMs. The proposed PP-ProPS is evaluated on both continuous and discrete control problems spanning Multi-Joint dynamics with Contact (MuJoCo) locomotion, classic control, highway driving, and robotic arm manipulation. Compared to Vanilla ProPS, the proposed PP-ProPS outperforms ProPS in seven of the ten evaluated tasks, and surpasses conventional RL methods including PPO, SAC, and TRPO, in five of the six tasks.

Abstract: Vision-language models such as CLIP achieve strong zero-shot classification, yet under distribution shift, visual embeddings drift from fixed text embeddings. Training-free calibration avoids the per-sample optimization of prompt learning, but prior feature calibration gives each image the full bias of one hard cluster. We propose Domain Recentering with Confidence Calibration (DRC), a training-free method adapting CLIP from a set of unlabeled target images. DRC fits a Gaussian mixture once and subtracts from each embedding a posterior-weighted average of component means. It then removes residual class preference with a log-prior correction, estimating the prior from confidence-weighted predictions. Among compared methods, DRC achieves the highest average accuracy on cross-domain datasets, exceeding zero-shot CLIP by 4.13 and 5.07 points with ViT-B/16 and ResNet-50, with gains over CLIP also holding under ImageNet distribution shifts.

Abstract: In value-based reinforcement learning, improving the accuracy of policy evaluation has been shown to improve downstream policy optimization performance. The widely adopted family of approximations relying on $n$-step truncation yields computationally efficient value estimators but is inherently limited to a short evaluation horizon. In contrast, methods that exploit the global structure of the transition dynamics can accelerate policy evaluation, but their memory and computational requirements often limit scalability to large or continuous state spaces. To reconcile these limitations, we introduce Limiting-Kernel Q($λ$) (LKQL), an off-policy value estimator that combines $n$-step truncation with a long-horizon approximation based on the limiting kernel (LK). LKQL has the same order of complexity as $n$-step estimators and integrates directly into both on- and off-policy actor-critic algorithms. We prove that, under aperiodicity and in the near-on-policy regime, the operator underlying LKQL improves the policy evaluation convergence rate over its truncated counterpart for sufficiently large $n$, and that LKQL itself converges almost surely to the optimal values in finite Markov decision processes (MDPs) under a fixed behavior policy. On the MuJoCo continuous-control benchmark, we show that LKQL improves over $n$-step baselines in most settings, particularly on long-horizon tasks.

Abstract: Embodied reinforcement learning (RL) improves model capabilities with a pipeline of environment simulation, action generation, and model updates. These stages show heterogeneous CPU and GPU demands, making efficient resource utilization difficult. Recent systems overlap rollout (simulation and generation) with training for efficiency, but exclusive GPU allocation and synchronized barrier in rollout still leave substantial hardware resource waste. In this paper, we present EBRL, an asynchronous embodied RL training system with two core techniques. The asynchronous pipelined scheduler overlaps rollout and training, pipelines simulation and generation across environment groups, and carries out each environment independently, eliminating synchronization stalls. The fine-grained resource manager pools CPU cores and GPU streaming multiprocessors, and uses stage profiles and runtime feedback to adjust resource quotas and batch sizes to meet the shifting demands among stages. We implement EBRL on RLinf and evaluate it with four embodied policies and four simulation benchmarks across heterogeneous GPU testbeds. Experiments show that EBRL achieves 1.30-3.47 times the end-to-end rollout throughput and 2.5 times of training convergency compared to the SOTA embodied RL systems.

Abstract: Unsupervised domain adaptation (UDA) for semantic segmentation remains challenging under adverse weather conditions because severe appearance changes enlarge the domain gap and degrade the reliability of pseudo labels in the target domain. To address this problem, we propose an Intra-Class Mixing Consistency (ICM) framework that enforces prediction consistency between an intra-class mixed image and its original counterpart. Unlike previous mixing-based consistency methods that combine regions across different images or domains and may introduce unrealistic semantic inconsistencies, ICM performs mixing within the same image and semantic class, preserving realistic semantic layout for consistency regularization. With ICM, we establish a new state-of-the-art performance for clear-to-adverse-weather unsupervised domain adaptation (UDA) in semantic segmentation. On the Cityscapes $\rightarrow$ ACDC benchmark, our method achieves 75.7\% mIoU, outperforming the previous state of the art by +1.9 pp, demonstrating its effectiveness in mitigating class confusion under challenging environmental conditions. The code is provided in the supplementary material.

Abstract: Camera localization in a prior LiDAR map provides a persistent geometric reference for long-term robotic navigation, yet remains challenging because of the substantial modality gap between camera images and point-cloud maps. We present a unified localization framework that makes the LiDAR map visually addressable rather than relying on a dedicated image-LiDAR correspondence model. Map geometry and reflectivity are rendered into LiDAR-derived quasi-images with explicit 2D-3D provenance, enabling camera observations and rendered map views to share mature visual features and matchers for both global localization and continuous pose tracking. Point and line correspondences are established through this common visual interface, while the retained provenance recovers metric LiDAR geometry and line-supported planar constraints for pose estimation. To improve robustness under ambiguous associations and weak geometry, we further introduce a distribution-aware, observability-complementary optimization strategy. Instead of reducing matching ambiguity to a scalar confidence, candidate association distributions are propagated into directional pose-information uncertainty, and reliable structural factors are selectively reinforced according to their ability to complement the currently weak pose directions. Experiments on the EuRoC MAV benchmark and self-collected real-world sequences demonstrate accurate global localization and robust continuous 6-DoF tracking using only a pre-built LiDAR map as the persistent prior, including under severe illumination variations and dynamic occlusions.

Abstract: Infrared (IR) imaging is crucial for autonomous driving, remote sensing, and other perception tasks. However, adverse weather may introduce fake structural responses that are entangled with real thermal structures. Existing IR restoration methods are typically designed for a single degradation type or directly reconstruct from degradation-entangled representations. Consequently, they struggle to distinguish intrinsic thermal structures from weather-induced fake responses and to accommodate spatially varying degradation severity, leading to artifacts or the over-suppression of weak but meaningful thermal responses. To address these issues, we propose TSGPD-IR, a type-severity guided progressive disentanglement network for all-in-one infrared restoration that factorizes restoration guidance into task-level weather semantics and region-level degradation severity. Specifically, a Weather and Semantic Co-Guided Multi-Level Prompt Generation Module combines global weather semantics with stage-wise local features to generate adaptive prompts that progressively suppress degradation-induced responses while preserving intrinsic thermal structures. To complement global weather semantics with spatial restoration control, a Proxy-Supervised Regional Degradation Estimator derives severity supervision without manual annotations and predicts spatially varying degradation priors. Guided by these cues, a Multi-Source Collaborative Expert Selection Strategy uses a shared branch to preserve weather-invariant thermal structures and hierarchical routing to select weather-specific expert pools and severity-compatible regional experts. This design progressively separates degradation interference from genuine thermal content and enables region-adaptive restoration, reducing both residual artifacts and over-suppression.

Abstract: Event cameras promise low-latency perception for high-speed robotic systems, where even short delays can render detections stale by the time they inform downstream robotic decisions. Yet modern event detectors still require tens of milliseconds of computation before their predictions become available. Conventional evaluation ignores this delay by comparing predictions with annotations at the observation timestamp, even though the scene may have changed by the time those predictions are produced. We study this observation-availability mismatch in event-based multi-object detection and show that state-of-the-art event detectors degrade substantially when evaluated at prediction availability rather than observation time. To address this, we introduce ChronoFuse, a causal availability-time detector that predicts object states for when its output becomes available rather than for when its input was observed. ChronoFuse performs causal cross-time fusion over a multi-scale feature hierarchy, combining current representations with cached temporal features to expose short-term temporal cues without using future observations. The fusion pathway is lightweight, adding only 0.17 million parameters and 0.84 ms of mean end-to-end latency overhead. ChronoFuse recovers 71% of the accuracy lost to latency on 1Mpx driving data and 90.8% under rapid drone motion on FRED, nearly restoring zero-delay performance. Under the extreme motion of EV-Flying, ChronoFuse reaches 20.95 sAP, compared with 2.25 for the strongest standard event detector (9.3x gain). These results show that predicting ahead can be critical for robots operating in fast-changing scenes, including autonomous driving, agile flight, and robotic interception.

Abstract: We study decentralized partially observable team decision problems with low-rank latent dynamics and unknown system models. The proposed framework combines team-theoretic equivalence with low-rank model representations to address cooperative decision-making in partially observable Markov decision processes without prior knowledge of the transition model. Each team member makes decisions based on local private information and delayed common information shared across the team. Using only this available information, each member learns an approximate low-rank Markov decision process and applies least-squares value iteration to compute its policy. This yields a fully decentralized learning and planning algorithm that requires neither a centralized coordinator nor centralized training. We show that the resulting member-side solutions approximate the centralized team solution: despite partial observability, unknown dynamics, and delayed common information, each member recovers the corresponding component of an approximate team-optimal policy. We further establish finite-sample performance guarantees and derive a corresponding sample-complexity bound for the proposed algorithm.

Abstract: When we evaluate the performance of our odometry, it is common practice to score the estimated track against a ground truth. Unfortunately, scoring uses point metrics, such as the root mean square error, that ignore the covariance matrix which estimators like filters and smoothers already report. Using the covariance matters for two reasons. First, the covariance encodes the estimator's uncertainty, so it tells us whether the estimator trusts its own output. An overconfident estimator will not report itself lost. Second, the covariance weights the error in each direction of the estimate. Without the covariance, an estimator is unduly penalized for a high error in an uncertain direction. Instead of point metrics, we should use strictly proper scoring rules. These rules score the estimate together with its reported uncertainty. Strictly proper scoring rules recover the point metrics when no covariance is reported, and they diagnose covariance inconsistency when covariance is reported. Using a one-sided pairwise test, we show that two estimators can expose overconfidence in at least one of them without a ground truth. Strictly proper scoring rules and our pairwise test are available in our open-source framework smfeval. As a case study, we use smfeval to assess the uncertainty quality of the translational component of ground-based LiDAR-inertial odometry. Across four filters we find overconfidence - the worst case reports centimeter certainty with kilometer error. Knowing the filters are overconfident, we investigate the mechanism. The investigation traces overconfidence to filters crediting LiDAR measurements with more new information than they carry.

Abstract: High-rate dense perception in dynamic environments is limited by the low update rate of RGB cameras, as rapid scene changes can occur between frames. Event cameras offer temporally dense but spatially sparse measurements, complementary to spatially dense RGB observations. Direct fusion cannot fully exploit this complementarity, while event-guided propagation fails on newly appearing or disoccluded regions without valid RGB support. We present LiFR v2, a unified propagation-completion-memory framework for causal anytime and streaming dense prediction from an RGB keyframe and events. LiFR v2 introduces an Event-Guided Completion Module (EGCM) to recover task-relevant representations where propagation is unsupported, and a History Retrieval Module (HRM) to reuse completed representations across successive queries. The framework supports semantic segmentation, monocular depth estimation, and multi-task dense prediction, and we further introduce SHF-Emerge to evaluate rapid object emergence and disocclusion. LiFR v2 achieves 74.37% mIoU on DSEC and 56.13% on SHF-Emerge, improving LiFR-Seg by 1.85 percentage points on the latter, while reducing SHF-Emerge depth RMSE from 1.564 m to 1.118 m over the propagation baseline. It also exceeds 100 FPS for both segmentation and depth, demonstrating accurate and efficient high-rate perception beyond RGB frame rates.

Abstract: LiDAR point clouds acquired in underground environments exhibit severe geometric incompleteness due to occlusions and limited sensor viewpoints, making reliable point cloud completion challenging without large supervised datasets. We propose ZVeC, a zero-shot, instance-driven framework that reformulates scene-level completion as compositional object-level reconstruction. By decomposing a scene into semantic object instances, ZVeC reduces reconstruction ambiguity in cluttered environments while eliminating the need for scenario-specific training. Each segmented vehicle is completed independently using a depth- and 3D Gaussian-conditioned diffusion model that exploits generalized geometric priors before the reconstructed instances are recomposed into the original scene. To evaluate our approach, we construct a real-world dense LiDAR benchmark of underground parking environments. Experimental results demonstrate consistent improvements over representative scene-level baselines in both quantitative metrics and visual quality. The completed point cloud differs substantially from the measured input (average KL divergence ~ 2.1), yet reducing the input to only 1% of the original LiDAR measurements changes the completed reconstruction only marginally (KL divergence < 0.50). This demonstrates that ZVeC produces geometrically consistent completions even under extreme input sparsity.

Abstract: Physical AI requires models to ground visual and linguistic understanding in real-world environments while accounting for environmental constraints and execution feedback. We introduce MachEmbodied-VLM (ME-VLM), a unified vision-language model with two variants, 4B and 35B-A3B, that brings together embodied cognition and multimodal agent capabilities. Our work emphasizes physical perception and spatiotemporal reasoning, together with planning, interaction, and outcome assessment in both digital and physical environments. We construct training data spanning embodied and multimodal agent tasks, including execution observations and feedback to support outcome assessment and decision refinement. The training pipeline comprises embodied capability injection, separate reinforcement learning of embodied and multimodal-agent experts, and multi-teacher on-policy distillation that consolidates their complementary capabilities into a single model. Experiments show competitive performance on both embodied and agent benchmarks, as well as on autonomous-driving and embodied-navigation tasks. For edge deployment, visual token compression, W4A8 quantization, and hardware--software co-optimization enable on-device inference of the 4B variant on the M100, reducing prefill latency from 400 ms to 188 ms. Project Page: https://machembodied.com/ME-Brain/ME-VLM.html Code Repository: https://github.com/MachEmbodied/ME-VLM

Abstract: Four-dimensional (4D) Radar has emerged as a key sensor for environmental perception, providing range, azimuth, elevation, and Doppler measurements while remaining robust to illumination changes and adverse weather conditions. However, conventional Radar preprocessing methods, such as constant false alarm rate (CFAR) detection, select measurements primarily based on signal-level criteria and may therefore discard information valuable for downstream perception during point cloud generation. In addition, existing 4D Radar perception pipelines typically optimize Radar data processing and downstream perception independently, preventing task objectives from directly guiding the preprocessing stage. To address these limitations, we propose a Scene- and Task-Aware Radar (STAR) Preprocessor together with an end-to-end training framework. The STAR Preprocessor incorporates scene context and downstream task objectives to generate task-relevant Radar points, enabling the Radar representation to be optimized directly for perception. On the K-Radar benchmark, the proposed method achieves 74.3 AP, outperforming the previous state of the art by 5.6 AP points. Furthermore, applying the task-relevant points generated by STAR to various existing 3D detectors improves detection performance in most evaluation settings and yields an overall positive average gain over point clouds produced by conventional preprocessing.

Abstract: Atomic activity understanding aims to recognize and localize structured traffic behaviors that jointly encode motion patterns and their grounding in road topology. Unlike conventional action recognition, atomic activities are multi-agent, multi-label, and topology-aware: multiple activities co-occur while many agents remain inactive. We introduce Action-Slot, a structured action-centric representation learning framework. Slot attention is widely used for object-centric decomposition, but its permutation-invariant design and object-level inductive bias are misaligned with atomic activity semantics. We reformulate slot learning as structured activity decomposition through three designs: (1) category-aligned action slots that anchor slots to predefined activity categories, (2) parallel spatio-temporal slot updating for holistic video-level reasoning, and (3) background and negative-slot regularization that enforces competition between foreground activities and irrelevant regions. Together these establish an activity-centric inductive bias that disentangles concurrent and asynchronous activities directly from raw video. Beyond recognition, the learned representations encode transferable spatio-temporal grounding signals. We further propose an attention-difference-based pseudo mask selection framework that suppresses false positives by measuring attention changes before and after candidate region removal, enabling weakly supervised localization without dense annotations. To support systematic evaluation, we introduce TACO, a balanced synthetic dataset with full atomic activity coverage and pixel-level annotations. Experiments on OATS, TACO, and annotated nuScenes show superior recognition, strong sim-to-real transfer, and state-of-the-art weakly supervised localization.

Abstract: Reinforcement learning for off-road navigation requires extensive vehicle-terrain interaction data, which are costly to collect in high-fidelity simulation. World models offer a promising alternative by replacing simulator roll-outs during policy optimization. However, an off-road world model must condition state transitions on exteroceptive terrain information, which proprioception alone does not provide. This challenge is further amplified by the need to model both rigid and deformable terrain, where data-driven and physics-based approaches offer complementary strengths. We propose Verti-WM, a physics-aided exteroceptive world model that recurrently fuses a frozen Transformer for rigid terrain and a neuro-symbolic terramechanics model for deformable terrain. Elevation and semantic observations queried from a supplied map at each predicted pose condition fusion, enabling six-degree-of-freedom rollouts for policy optimization without further simulator access. Verti-WM reduces prediction error by 34.6% and 21.7% over data-driven and physics-based baselines, respectively. Policies trained entirely within Verti-WM achieve comparable task success rates while reducing computation time by 23.6X relative to direct training in the high-fidelity simulator. We further validate Verti-WM using real-world data, enabling policy optimization within learned real-world kinodynamics and achieving a 80% success rate on the Verti-4-Wheeler platform, compared with 40% for direct sim-to-real transfer.

Abstract: Multi-agent unmanned systems are moving from isolated, ego-centric sensing toward collaborative intelligence, in which distributed agents exchange compact features to overcome a local observation trap that no single agent can escape: occlusions, finite sensor range, and environmental degradation. The field has matured across architectural, communication, embodied, resilience, and trust dimensions, yet existing surveys examine these dimensions in isolation and rarely expose their dependencies. This review offers a unified synthesis through two complementary lenses. The first is a five-dimensional taxonomy spanning collaboration stage, communication paradigm, fusion architecture, learning strategy, and application domain. The second is three cognitive synergy conditions, Semantic Disambiguation, Pragmatic Information Exchange, and Proactive Informational Foraging, that turn cognitive synergy into operational criteria. Across these lenses we survey collaboration architectures and topologies, neural-communication co-design that treats the channel as a differentiable pipeline component, embodied action-perception loops via multi-agent reinforcement learning, and resilience mechanisms for synchronization, uncertainty quantification, and label-efficient learning. We then map these advances onto four operational domains, V2X, unmanned aerial, industrial logistics, and smart cities, and onto the safety-privacy-utility triad. To counter benchmark saturation and evaluation fragmentation, we propose GCI-Bench, a five-pillar scoring protocol with a maturity model that makes the trade-offs of collaborative methods comparable across studies. A critical reflection on reproducibility, the sim-to-real gulf, and conditions under which collaboration degrades performance identifies open challenges and charts directions toward general collaborative intelligence under real-world uncertainty.

Abstract: Quadrupedal robots are increasingly deployed in environments where locomotion must remain robust to disturbances and constrained terrain. Gait type, such as walking or trotting, is commonly used to characterize quadrupedal locomotion. However, gait type does not uniquely define locomotion, as parameters such as duty factor, speed, and stance width can vary within a single gait type. In this work, we investigate the relationship between these gait parameters using three distinct quadrupedal locomotion control approaches. First, using whole body trajectory optimization with LQR feedback, we show that duty factor is a stronger predictor of local error convergence than nominal gait type. Second, we investigate duty factor selection with a learned locomotion controller, suggesting how duty factor may serve as a low-dimensional parameter for adapting locomotion robustness in narrow-terrain environments. Finally, we show that these trends persist under a centroidal model predictive control framework and validate them through narrow-terrain experiments on a physical quadruped. These results show that duty factor provides a simple and effective basis for understanding and selecting robust quadrupedal locomotion across gait types and control architectures.

Abstract: Behavioral failures can make a transformer appear to lack a world model even when it has learned faithful representations of its environment. We demonstrate this in TaxiGPT, a transformer trained on random walks through Manhattan whose failures have been interpreted as evidence of an incoherent internal map. Through mechanistic analysis and causal interventions, we show that the model represents intersections and streets, tracks its position, and uses a goal compass to navigate. We trace its failures to interference between superposed intersection features, which disrupts localization within the internal map. Affordance packing, which groups representations of intersections with the same legal moves, helps limit the consequences of these errors. Finally, we propose mechanistic indicators that we use to compare models and show that world-modeling capacities emerge at different stages of training. Our findings motivate a shift from asking whether a model has a world model to mechanistically studying its world modeling: the interacting capacities through which it represents its environment and uses those representations to guide behavior.

Abstract: A robot can predict failure and still be unable to prevent it. By the time a safety mechanism reacts, the nominal plan may already have spent the control authority that recovery requires, and fixed task priorities may block whatever response remains. Our key insight is that both aspects are decided inside the controller. Recoverability must inform actions while they are chosen rather than veto them afterward, and task objectives must be adapted as recoverability shrinks. Building on this, we present RAYA, a hybrid learned-analytic framework that places a learned finite-horizon recoverability margin inside an optimal controller with hard constraints and pairs it with a bounded learned scheduler that shifts task weights to facilitate recovery. Across 7,200 simulation episodes per controller spanning quadrotor and autonomous-vehicle benchmarks, RAYA not only improves survival rates, but also transfers the learned components zero-shot to unseen trajectories, disturbances, plant shifts, and friction layouts. We developed an embedded realization of RAYA and deployed it on-board a 35g Crazyflie quadrotor. Across 40 combined hardware flights under wind with either aerodynamic mismatch or an unmodeled 40% motor-command loss, each of three baselines fails in all trials, while RAYA completes 10/10 six-cycle missions. Project Website: https://raya-control.github.io/.

Abstract: Substantial efforts have been devoted to making deep learning objectives, representations, and architectures interpretable, with the goal of improving the safety, robustness, and generalisation of learning systems in diverse real-world applications. The recently proposed maximal coding rate reduction ($\mathrm{MCR}^{2}$) offers a promising information-theoretic framework for learning structured, discriminative representations of class-wise submanifolds and has inspired interpretable white-box architectures. However, we observe that $\mathrm{MCR}^{2}$ can completely fail under distribution shift, motivating our study of its out-of-distribution (OOD) generalisation limits. We establish two limitations of $\mathrm{MCR}^{2}$ for OOD generalisation. First, the $\mathrm{MCR}^{2}$ objective alone can admit complete prediction failure: a representation based entirely on unstable environmental features can achieve the global coding optimum yet fail completely after correlation reversal, despite an available perfectly stable feature. This exact-optimum example includes test inputs that cannot occur during training. Even when every possible test input can also occur during training, coding quality can be arbitrarily close to optimal while prediction error is arbitrarily close to 100%. Second, directly incorporating the invariance principle underlying widely successful invariant risk minimisation (IRM) and risk extrapolation (REx) does not eliminate this failure. The failing representation admits the same optimal coding operator across training environments, showing that shared coding optimality does not ensure stable prediction. Reliable OOD guarantees for $\mathrm{MCR}^{2}$ therefore require additional new assumptions or learning principles that establish stable predictive relationships across environments.

Abstract: Cross-view object geo-localization (CVOGL) typically assumes a fixed query image, overlooking the ability of mobile agents to actively acquire more informative observations. To address this limitation, we introduce Active Cross-View Object Geo-Localization (ActiveGeo), where an agent sequentially selects new viewpoints and determines when to stop, aiming to improve localization with minimal observations. We further propose ActiveMoPT, an ActiveGeo framework with three-stage training. First, Multi-View Prompt-Preserving Adaptation enables the model to aggregate multiple query views while reusing the initial prompt. Second, Trajectory-Guided Policy Initialization uses supervised agent trajectories to learn viewpoint selection and initial stopping behavior. Third, Cost-Aware Policy Refinement employs GRPO with a gain-cost reward to jointly optimize localization accuracy and observation efficiency. We also construct ActiveGeo-858, a zero-shot test set containing 858 scenes and 1,716 target annotations. Experiments show that ActiveMoPT achieves state-of-the-art performance on MoP-UAV using only 1.45 query views on average, and substantially outperforms previous CVOGL approaches under zero-shot evaluation on ActiveGeo-858.

Abstract: We propose SEAM, a LiDAR-based lifelong mapping framework. Instead of relying on a single anchor spanning the entire session, SEAM generates evidence based on a trajectory optimized with submap-level anchors, and performs dynamic object removal and change detection. Through submap-level reprojection, the generated evidence remains usable even if the trajectory is subsequently modified by a new session, eliminating the need to recompute the entire process from scratch. SEAM suppresses geometrically unreliable inter-session loop edges using a DOP-based confidence measure. Suppressing unreliable loop edges prevents alignment errors. SEAM also uses a directional voxel-wise evidence model. The model accounts for occupancy patterns that vary with ray direction. Direction-aware evidence separates dynamic objects from environmental changes more precisely. Experiments on a real construction-site dataset and a long-term multi-session dataset show that SEAM achieves higher accuracy and faster processing than existing methods.

Abstract: Vision-Language Models (VLMs) can produce Natural Language Explanations (NLEs) that sound plausible yet remain inconsistent with the visual evidence they cite. We present Explanation-Driven Counterfactual Testing (EDCT), an intervention-based protocol that extracts visual concepts cited in a model's explanation, applies verified minimal edits to them, and tests whether the resulting answer and explanation remain consistent with the edited image. Using this protocol, we create EDCT-Bench, a comprehensive benchmark spanning three complementary domains: knowledge-intensive visual question answering (OK-VQA), safety-critical driving (DriveLM), and 3D spatial reasoning (3DSRBench). Across the evaluated VLMs, EDCT reveals substantial faithfulness gaps, with models frequently producing responses inconsistent with verified visual changes. Finally, our fine-tuning study suggests that EDCT-generated counterfactuals provide high-impact training signals.

Abstract: Image segmentation remains challenging due to occlusions, poor lighting, and irregular structures. Although transformer-based methods achieve high accuracy, they rely heavily on long-range spatial features, leading to high computational costs and neglecting prior knowledge or noise patterns, resulting in missing details and unclear boundaries. To address these issues, we propose Frequency Spatial Aware Network (FSANet), which integrates prior knowledge with a dual-domain solver to sequentially adapt to diverse segmentation tasks. Specifically, we design three key modules: (1) Structure Prior Module, which recovers overlooked details; (2) Dual-Domain Awareness Module, which captures salient features while disentangling noise; and (3) Edge Estimation Module, which enhances edge awareness for more precise segmentation. In addition, the limited availability of comprehensive segmentation datasets covering various real-world scenarios hinders the performance of existing methods. To address this, we introduce SceneX, a novel open-source dataset featuring 10 challenging non-ideal scenarios, establishing a new benchmark for evaluating and improving the robustness and real-world applicability of the segmentation models. Extensive experiments demonstrate the efficiency and effectiveness of FSANet.

Abstract: Large-scale pre-training has transformed representation learning in 2D vision, yet its transferability to 3D single object tracking (SOT) remains insufficiently understood. Directly fine-tuning self-supervised 3D encoders, such as masked autoencoders (MAE), often leads to sub-optimal adaptation because the reconstruction objective is not fully aligned with the spatial-temporal matching requirements of tracking. In this paper, we observe that this difficulty can be interpreted as a layer-wise transfer mismatch: shallow layers tend to preserve transferable geometric cues, while deeper layers become increasingly specialized to the reconstruction pretext task and are less suitable for downstream tracking. Based on this observation, we propose MAETrack, a lightweight adaptation framework for transferring pre-training MAE representations to 3D SOT. MAETrack includes Layer-Selective Initialization (LSI), which initializes only the shallow stages of the tracking backbone from pre-trained weights while re-initializing deeper stages, and Geometric Residual Gating (GRG), which reinforces structurally salient regions in the search BEV features before template-search fusion through residual spatial modulation. Extensive experiments on standard 3D SOT benchmarks show that MAETrack consistently improves upon vanilla fine-tuning baselines with limited computational overhead. More broadly, our results suggest that effective transfer from 3D reconstruction pre-training to 3D tracking is not merely a matter of partial fine-tuning, but depends on a tracking-oriented transfer principle that preserves shallow geometry while adapting deeper representations to the downstream objective.

Abstract: Stereo cameras are integrated into autonomous systems such as self-driving cars, drones, and robots to offer precise depth estimation in a cost-effective manner compared to LiDAR technology. In this work, we reveal an intrinsic vulnerability in stereo cameras that stems from their pixel sampling and calibration processes, which can influence the outputs of stereo matching algorithms. Attackers can achieve fine-grained control over the estimated depth of real obstacles using simple repeating patterns, without relying on sophisticated adversarial machine learning techniques. Furthermore, deep learning-based depth estimation models exhibit a similar vulnerability. We evaluate the impact of this attack on two widely used stereo matching algorithms (BM and SGBM), three deep learning models (PSMNet, MoCha-Stereo, and UniMatch), a stereo-LiDAR fusion model (SGM-DDC), and two popular commercial stereo cameras, the ZED2 and Intel RealSense D435. For example, in the ZED2 camera, an attacker can displace obstacles up to 20~meters farther or 12~meters closer. In our real-world evaluation in a driving setting, a brief 0.5~second attack can trigger emergency braking in a popular autonomous driving framework. We further demonstrate the feasibility at driving speeds up to 40~km/h using CARLA. Finally, we confirm the ineffectiveness of state-of-the-art defenses, and we propose a novel strategy that leverages similarity scores to dynamically detect and suppress the depth discrepancies. Our work highlights vulnerabilities hidden in stereo matching and deep learning depth estimation models, addressing critical limitations in autonomous system deployments.

Abstract: Vision-language-action policies are typically evaluated one perturbation at a time, providing a useful diagnosis of their sensitivity to individual distribution shifts. Real-world deployment, however, may involve several shifts simultaneously, and it remains unclear how these individual robustness measurements compose. We ask whether compound robustness can be inferred from single-axis evaluations. We introduce LIBERO-CTRL, a six-axis benchmark that pairs each initial state across single-axis conditions and a matched simultaneous condition. This design reveals two opposing outcome changes that aggregate success rates cannot distinguish: emergent failures, where all single-axis rollouts succeed but the simultaneous rollout fails, and compensated successes, where at least one single-axis rollout fails but the simultaneous rollout succeeds. Because one transition decreases compound success while the other increases it, they can cancel, making aggregate compound performance appear consistent with single-axis measurements even when individual outcomes differ substantially. These opposing transitions can largely cancel in aggregate: even when the difference between the two transition rates is not statistically distinguishable from zero, as many as 29.0% of matched initial states still change outcome. Across six policies and three severity levels, such outcome changes reach 34.5% in the most affected condition. The relative prevalence of the two transitions varies across policies and severities, while the transition rates remain similar under independent re-evaluation of stochastic policies. Compound robustness therefore cannot be characterized from aggregate single-axis success rates alone; matched per-instance evaluation is needed to reveal how joint perturbations alter behavior.

Abstract: A robot following language instructions needs its semantic memory to keep naming the same physical object while the SLAM pose graph underneath is optimized, loop-closed and compressed. Maps committing each detection to a world coordinate cannot: a closure moves the anchor it was measured from, or the solver marginalizes that anchor, and the query then selects a different object although both graphs represent the same posterior. P-POSEMEM stores each observation as an immutable event at its birth keyframe, retains the Bayes-tree elimination conditional of every marginalized keyframe, and integrates the semantic likelihood over the reconstructed joint posterior of poses, anchors and identities. Dproj, the total-variation defect between the language-goal distributions of inference-equivalent full and marginalized graphs, measures this directly. Over 40 HM3DSem scenes and 112,000 queries, P-POSEMEM reproduces the full-graph oracle (Dproj = 0) and reduces goal flips against every memory-reducing baseline. On an eight-run campaign whose 761 closures rewrote the map by up to 47 m, Dproj stays below 10^-13 with 0/288 goal flips when elimination follows the closures, where every ablation and a coordinate committed at insertion flip goals it does not; under a live bounded solver the same memory flips 23/288 against 53 for that frozen coordinate. A pre-registered negative control is detected by Dproj while leaving calibration error and navigation success unchanged, indicating that these measures capture distinct failure modes. Retrieval is held fixed by a shared frozen detector, isolating the gain to memory consistency. Code and data: https://anonymous.4open.science/r/posemem-2328/.

Abstract: Safe reinforcement learning (RL) is commonly formalized as a Constrained Markov Decision Process (CMDP), in which an agent maximizes expected reward while keeping its expected cumulative cost below a specified safety bound. Existing safe RL benchmarks predominantly report whether an algorithm is safe on average, following this expectation-based guarantee. We argue that this convention is insufficient to reliably characterize an algorithm's true safety: it fails to capture how often and how severely the safety bound is violated, whether this holds consistently across tasks and safety bounds, and whether training-time behavior is representative of behavior of the final converged policy. Therefore, we introduce (i) evaluation metrics for safe RL that address each of these concerns and in addition allow for aggregation across tasks and safety bounds. We furthermore define (ii) a safety tier system to systematically categorize and compare algorithms in terms of safety and reliability at both training and for a final policy. Using this framework, we provide (iii) an empirical safety evaluation across multiple safety navigation tasks. Our results show that aggregate metrics, distributional reporting, and task- and safety bound-specific results each reveal information the other metrics cannot. We therefore recommend reporting all three jointly, rather than compressing this information into a single value, as is common practice. We provide SafeRLEval, an open-source evaluation suite to support the reliable characterization of safety in future safe RL research.

Abstract: Accurate 3D localization of unmanned aerial vehicles (UAVs) remains challenging for existing multimodal approaches due to sparse LiDAR geometry, modality-imbalanced fusion, and redundant feature transmission over constrained edge-to-server links. To address these limitations, we propose PRI-Net, an efficient and lightweight multimodal fusion framework for UAV localization that integrates point cloud splatting, residual attention fusion, and an information bottleneck. Specifically, a 3D point cloud splatting (3DPCS) strategy is introduced to transform sparse LiDAR observations into geometrically consistent dense depth maps. A residual attention fusion (RAF) module is then designed to alleviate modal bias by using an image branch for coarse estimation and a gated fusion branch for refinement. In addition, a multimodal information bottleneck (MIB) module compacts features by filtering task-irrelevant redundancy. Experiments show that PRI-Net achieves high localization accuracy with lightweight architectures, while reducing feature dimensionality and improving edge-to-server UAV sensing efficiency and robustness.

Abstract: As the use of Large Language Models moves from chatbots into agentic systems, where outputs become actions with irreversible consequences on reality, the existing paradigm on AI Interpretability research, post-hoc interpretability, is structurally inadequate for safe and trustworthy model deployment: it explains behavior after the fact but cannot audit or intervene in an inference computation before it commits to an output. We therefore argue for a shift toward \emph{generative interpretability}, an architectural property under which a model's inference pass natively exposes semantically meaningful checkpoints that are human-understandable and amenable to causal intervention. We show the merits of generative interpretability as comparison to other interpretability research paradigms, and propose Neuro-Symbolic Models as a concrete instantiation.

Abstract: A world model asked what happens if an object is pushed at several speeds produces several futures. If the model has the object in mind, those futures agree about it: each implies the same mass and friction. The consistency checks now used to vet world-action models ask whether a model's futures agree with each other, and none of them knows any physics. We show that this is not enough, and what to do instead. IMPLY reads the physics each rollout implies by inverting a simulator and scores a set of rollouts by how well one object explains all of them, anchored to two calibration pushes the model has observed. In a controlled setting, self-consistency gives a perfect score to a model that ignores the object and always predicts a typical push; anchoring exposes it (AUROC 0.70 versus 1.00). On a real model, V-JEPA 2-AC adapted to the scene, the same thing happens. Given its own calibration pushes the model tracks the object (per-object correlation with the truth 0.91); given another object's, it does not (0.05). Self-consistency cannot tell these apart, preferring the right evidence on 52% of objects, chance level, while anchored disagreement prefers it on 73% and correlates 0.92-0.99 with the rollouts' error. Used to choose among candidate rollout sets, it comes within 0.003 of an oracle that sees the truth. A model that has internalised the wrong object is exactly as self-consistent as one that has internalised the right one; consistency has to be anchored to evidence.

Abstract: Maintaining consistency over long spatial and temporal horizons remains a fundamental challenge in large-scale LiDAR SLAM, particularly when integrating maps collected across multiple sessions. We present Chain-SLAM, a LiDAR SLAM backend enabling online multi-session map alignment and reuse with global consistency at large scale. We implement a chained loop closure mechanism that efficiently propagates geometric constraints across inter-session keyframes through an adjacency graph, enabling robust long-horizon consistency triggered by reliable short-horizon loop closures. The system initializes inter-session alignment with GNSS-proximity place recognition, then performs on-the-fly loop closure detections and joint optimization of loaded maps and newly acquired trajectories within a unified factor graph, maintaining both inter- and intra-session geometric consistency without dynamic object removal, and cross-platform robustness with minimal hyperparameter tuning. Experimental results show improved trajectory accuracy and robust multi-session integration on large-scale datasets. We release our source code to support reproducible research in large-scale multi-session LiDAR SLAM. Project site: https://ai4ce.github.io/Chain-SLAM/

Abstract: UAV video question answering requires separating camera motion from changes in the scene, but RGB-only multimodal models receive no explicit, stable reference for that separation. We present EgoSIS, a pose-free adapter that converts RGB-derived bidirectional flow into motion-canonical visual evidence in three stages. Factorized Visual Ego-Transitions (FVET) fits a robust image-plane transition and exposes motion, residual-support, and reliability factors. Reliability-Gated Ego-Transition Memory (ReTEM) uses reliability-weighted updates for a bounded history and re-anchors it at cuts or sustained uncertainty. Ego-Aligned Spatial Evidence (EASE) warps supported visual features into each segment's local anchor and injects four spatial evidence tokens per visual slice through zero-initialized residuals, without changing Qwen's visual-token count. On SIS-Bench, EgoSIS-8B obtains 89.9\% perception, 82.5\% perception-plus-memory, and 76.2\% overall accuracy, with the largest gains concentrated in self-awareness perception and memory. The adapter thus provides an interpretable interface between optical flow and spatial reasoning.

Abstract: Embodied planning increasingly relies on vision-language models (VLMs) to translate instructions and visual observations into executable action sequences. However, fluent plans are not always executable. A VLM may refer to objects that are not visually observed, select actions whose required affordances are unavailable, or violate syntax and action constraints. We introduce CLAMP, a multimodal constraint-grounding framework that turns scene evidence into decoding-time constraints for a frozen VLM planner. CLAMP uses the initial observation to restrict object references to those supported by the scene, while a provided symbolic action model specifies state transitions and goals. During decoding, hard masks eliminate invalid next-token candidates, while a Hidden Markov Model (HMM)-based world-state lookahead module reweights the probabilities of the remaining feasible candidates based on action preconditions and goal reachability. This allows the planner to retain the VLM's language prior while preventing visually unsupported, unsafe, or infeasible candidates from entering the plan. For unseen tasks and environments, CLAMP adapts the HMM at test time using label-free continuations sampled from the frozen VLM. Experiments on VLABench, SafeAgentBench, and TaPA show that scene-grounded constraints improve object grounding and safety, while most remaining failures stem from perception errors or misaligned constraint specifications.

Abstract: How can vision-language-action (VLA) models adapt to new environments where world dynamics shift? While recent research has combined world modeling and action prediction to improve VLA performance, existing methods largely rely on pretraining on static datasets, without mechanisms for active adaptation at deployment time. As a result, these models often fail to generalize when deployed in unseen scenarios with novel object configurations or dynamics. We present WorldAgen, a unified framework that jointly learns world modeling and action prediction while enabling Test-Time Training (TTT) to adapt to new environments. WorldAgen employs a shared Transformer backbone with two heads: (1) a world model head that predicts future states from past state-action trajectories, and (2) an agent model head that predicts actions conditioned on task instructions. We design a Mixed Unidirectional Attention Mask to separate these two models. During test time, WorldAgen samples exploratory actions, collects ground-truth state transitions, and performs lightweight TTT updates to refine its world model. This adaptation improves the model's understanding of the environment and leads to more accurate action predictions. Experiments on the CALVIN and LIBERO benchmarks demonstrate that our baseline model achieves comparable, and in some cases superior, performance to current state-of-the-art approaches. Moreover, with TTT on a small number of samples, our method surpasses existing state-of-the-art models, highlighting the effectiveness of adapting world models at inference time.

Abstract: Image restoration is commonly applied before object detection under adverse conditions, yet a visually improved image need not improve the downstream task. We study this mismatch as restoration utility prediction: given a degraded image and its candidate restoration, should the restoration be used or should the original observation be preserved? We introduce TaskGuard, a post-hoc controller for frozen restoration and detection pipelines. TaskGuard characterizes the realized restoration residual through its interaction with detector sensitivity and predicts whether the intervention is task-beneficial. Exact regional counterfactuals reveal substantial within-image utility heterogeneity, while a deployable pseudo-gradient preserves statistically reliable directional information. Feature-group ablation further shows that task-conditioned evidence contributes information beyond detector-response and residual statistics. The TaskGuard utility predictor is trained only on Gaussian degradation and frozen before final evaluation, then transferred to unseen motion blur, rain, and defocus. Across these unseen families, TaskGuard reduces lossnegative interventions by 54.2% (family macro) and practical per-image detection deteriorations by 37.0% (pooled), while preserving 98.8% of the Always-Restore COCO AP. On natural-rain DAWN, it reduces loss-negative interventions by 97.9% while retaining 77.8% of the AP improvement obtained by deraining. These results support restoration utility as a task-conditioned property of the specific intervention rather than image appearance alone.

Abstract: Vision-language models are increasingly used in settings where some input modalities may be unavailable, yet we know little about whether they can faithfully explain how such missing information affects their own predictions. We introduce an interventional protocol for evaluating self-explanations of modality dynamics: models state what each modality alone would support, whether restoring a missing modality would change their answer, and whether the available evidence is sufficient; we then execute the corresponding modality intervention and compare these claims with the model's realized behavior. We evaluate eight open-weight VLMs from two model families across four tasks spanning complementary and isomorphic text-image settings and a multi-view driving setting. We find a systematic tendency to overstate the sufficiency of available modality evidence. Models substantially underestimate the effect of restoring missing modalities: task-level median predicted change rates are at most 8.8%, while the corresponding executed change rates reach 72.1%, with underprediction in 62 of 64 model-task-condition settings. Insufficiency claims are rare, but precise when produced: restoring the modality changes the answer in a median of 78-100% of flagged cases. Retrospective self-explanations show the same tendency: on complementary data, models over-credit single-modality sufficiency; on isomorphic data, they over-credit single representation sufficiency relative to their executed behavior. Together, these results show that VLMs systematically mischaracterize how their predictions depend on available and missing modality evidence, motivating executable interventions as a behavioral ground truth for evaluating multimodal self-explanations.