Papers

Recently, Vision-Language-Action (VLA) models have revolutionized robotic manipulation by seamlessly integrating visual perception, language understanding, and action generation in an end-to-end learning framework. However, since these models are designed to interact directly with the physical world and humans, their security is critical, and even small vulnerabilities can lead to catastrophic failures. In this work, we propose the Universal Adversarial Object, a sphere with optimized surface texture that significantly degrades task success rates when placed within the robot's field of view. Specifically, our approach introduces a multi-level attack framework that jointly disrupts trajectory planning, task execution, and action control. We validate our method in both simulated and real-world robotic settings. Experimental results demonstrate that the adversarial object reduces the average task success rates by 31.2%-39.9% for two representative VLA models (Pi0 and RDT), with success rates dropping to near zero in complex scenarios. Index Terms–Vision-Language-Action models, adversarial attack, robotic security, universal adversarial object

Deep neural networks tend to rely on simple features that may be spurious and thus fail to generalize. We study this problem in the setting of linear probes, where a (generalized) linear model is fitted on the representations of a (pretrained) model. We use the connection of these models to the max-margin classifier, and show they favor directions associated with large eigenvalues of the covariance matrix. Whitening removes this preference by equalizing the eigenvalues of the covariance matrix. This observation motivates whitening as a preprocessing step that can reduce reliance on spurious correlations without requiring prior knowledge of their presence or labeled data. We examine the effect of whitening on a synthetic data-generating process and standard spurious correlation benchmarks, and find that it improves robustness. We also find that whitening can improve robustness when added to existing approaches.

Gausslet orbitals on a cubic lattice, introduced in [Steven R. White, J. Chem. Phys. 147, 244102 (2017)], represent a localized, smooth, and systematically refineable basis set featuring a "diagonal" approximability of the electron repulsion integral tensor. The present work develops efficient algorithms for evaluating overlap integrals required for electronic-structure simulations using these Gausslets, specifically the kinetic, nuclear, and electron repulsion integrals (both with and without the diagonal approximation). Computational efficiency improvements rest on the exploitation of translation, permutation, and octahedral symmetries, a reordering and precomputation of nested sums, as well as an early truncation of small coefficients. Our algorithms reduce the number of electron repulsion integrals on a $5 \times 5 \times 5$ grid from $125^4 = 244140625$ to $138187$ due to symmetries, and achieve a wall-clock runtime for evaluating the remaining integrals with a truncation tolerance of $10^{-5}$ in under 1 second on a laptop computer. We apply the developed methodology to compute the ground state of the hydrogen atom and molecule as a demonstration.

We report the temperature and pressure tunability of an unusual structural transformation associated with a two-step metal-insulator-metal (MIM) transition in the kagome lattice compound PrIr$_3$B$_2$ using synchrotron X-ray powder diffraction. At ambient conditions of temperature and pressure, the monoclinic ($C2/m$) and the hexagonal ($P6/mmm$) phases coexist as twinned structure in the crystal. As the temperature (pressure) is decreased (increased), PrIr$_3$B$_2$ converts fully to monoclinic structure at $T =$ 280 K (at ambient pressure) and $P =$ 1.2 GPa (at room temperature). Temperature dependence of the monoclinic structure at ambient pressure presents complex evolution of lattice parameters with weak but clearly discernable anomalies at \SI{\sim 250} {\K} and \SI{\sim 110} {\K}, which are correlated with the second MIM transition and the linear to nonlinear temperature-dependent resistivity crossover, respectively. These anomalies are likely due to some charge order state causing a partially gapped Fermi surface. The magnetic phase diagram of PrIr$_3$B$_2$ is also investigated from anisotropic measurements. At 10 K, a superzone gap opens near the antiferromagnetic transition, which does not close even in the polarized state. From the tunability of the crystal structure and magnetic and electronic ground state, promising electronic orders are indicated in this kagome metallic magnet.

This paper investigates electron transport properties across N/AM interface or N/AM/N heterojunction. Based on Hamiltonians in the three regions, general wave functions are proposed and combined linear equations are constructed. Reflection and transmission coefficients as well as spin polarizatioin of the altermagnet heterostructure are obtained. When electron incident from normal metal to altermagnet, it is scattered and split into + branch and _ branch transmitting into altermagnet zone with separate refractive angles. Total internal reflection might happen to + branch. Critical angle of total reflection is determined by incident conditions and properties of AM material. In the heterostructure case, spin-up and spin-down electrons transmitted with same propagation directions but different transmission probabilities. Transmission and polarization periodically change with length of the junction. Exchange field and spin-splitting strength have pronounced effect on spin-polarized transmittance while the effect of spin-orbit coupling is quite small. Optimized system can be found by tuning parameters of the AM material. This study will lay the theoretical groundwork for experimental research on spin-selective outputs and spintronic devices.

In this paper, we consider asymptotic properties of the support vector machine (SVM) in high-dimension, low-sample-size (HDLSS) settings under a spiked model. The existing theory of the SVM in the HDLSS context relies on the geometric representation of HDLSS data, which requires that the eigenvalues of the covariance matrices are not dominant. We first show that the geometric representation does not hold under the spiked model. We show that the Gram matrix of HDLSS data converges in distribution to a random matrix, namely, the HDLSS data converge to a random configuration in a finite-dimensional space whose dimension is given by the number of the spikes. We show that the misclassification rates of the SVM do not tend to zero, that is, the SVM does not hold the consistency property. We also show that the bias-corrected SVM (BC-SVM) does not give preferable performance in this setting because the bias term itself should be modified. In order to overcome such difficulties, we propose a spike-corrected SVM (SC-SVM). We show that the SC-SVM holds the consistency property when the sample size goes to infinity, and that the growth of the sample size is essential in the sense that any projection-based procedure fails when the sample size is fixed. Finally, we check the performance of the classifiers by numerical simulations.

Fingerprint recognition is a widely deployed biometric, but supervised training requires large labeled enrollment sets. Self-supervised learning (SSL) removes this requirement, and hybrid quantum-classical models have been proposed to enrich the learned representations. Prior quantum SSL studies consider a single contrastive objective, so it is unclear whether reported benefits depend on the objective or can be attributed to the quantum circuit. We insert the QuFeX quantum feature-extraction module into three SSL frameworks, the contrastive SimCLR and MoCo v2 and the non-contrastive BYOL, and compare each hybrid with its classical counterpart at matched representation width (8 features, equal to 8 qubits) on the SOCOFing fingerprint dataset, with a CIFAR-10 control, using k-nearest-neighbor identification on encoder features. In single-run experiments the hybrid scores clearly higher for both contrastive objectives, whereas for BYOL a multi-seed analysis shows no reliable difference, suggesting that any benefit depends on the SSL objective. A hardware-efficient circuit (QNet) does not show the same gain. We examine whether the gains can be attributed to the quantum circuit, considering circuit architecture, trainable parameter count, nonlinearity, and the classical simulability of 8-qubit circuits.

In this paper, we introduce a new set of $\mathbb{L}_2$-orthogonal polynomials on the hemisphere $\mathbb{S}^d_+:=\{\mathbf{x}\in\mathbb{R}^{d+1}:\|\mathbf{x}\|=1, \mathbf{x}\cdot\mathbf{e}_{d+1}\geq0\}$ and establish $\mathbb{L}_1$ Marcinkiewicz-Zygmund inequalities on the hemisphere for even and odd spherical polynomials. Hemispherical $t$-designs provide equal weight cubature rules on the hemisphere which are exact for polynomials up to degree $t$. We propose variational characterizations of hemispherical $t$-designs by the $\mathbb{L}_2$-orthogonal polynomials. Moreover, based on the Marcinkiewicz-Zygmund inequalities and the $\mathbb{L}_2$-orthogonal polynomials, we prove that for each $n\geq C_d t^d$, there exists a hemispherical $t$-design on the hemisphere with $n$ points, where $C_d$ is a constant depending only on $d$.

Nonlinear partial differential equations (PDEs) underpin computational fluid dynamics, yet resolving nonlinear transport on fine grids remains computationally demanding. Quantum linear-evolution algorithms offer a possible route to large-scale simulation but cannot directly propagate nonlinear coupling. Here we develop a Frozen–Oseen quantum homotopy method (FOQHAM) framework that links transport-aware auxiliary-operator selection to the size of the resulting linear representation. We freeze the full Fréchet derivative at a prescribed flow profile, retaining transport and profile-gradient coupling and close each prescribed finite-order homotopy hierarchy as a fixed affine linear system by product lifting. We formulate its propagation through Linear Combinations of Hamiltonian Simulations (LCHS), without outer homotopy iterations or profile updates. For spatially semidiscrete equations, we establish local approximation bounds and exact closure of the finite-order hierarchy. Classical tests on Burgers, Korteweg–de Vries, Zakharov–Kuznetsov, and one- and two-dimensional isothermal compressible Navier–Stokes equations provide numerical evidence of accurate non-iterative approximations in the tested regimes. The framework thus connects nonlinear PDE approximation to quantum linear evolution and offers a conditional path toward quantum acceleration.

In this paper, we establish Edgeworth expansions of extreme eigenvalue distributions for two types random unitary ensembles at the spectral edges and reveal certain universal structures for the correction terms. More precisely, we show the correlation kernels for the Gaussian-type unitary ensembles and Laguerre-type unitary ensembles admit full expansions at the soft edge and the hard edge, respectively. The coefficients of the correction terms are given by finite sums of Airy function or the Bessel functions of the first kind and their derivatives with polynomial coefficients. By lifting the kernel expansion to the associated Fredholm determinant, we further obtain full expansions for the largest eigenvalue distribu- tion of the Gaussian-type unitary ensembles and the smallest eigenvalue distribution of the Laguerre-type unitary ensembles. In addition, the first correction term therein involves the derivatives of the leading term.

Reinforcement Learning with Verifiable Rewards (RLVR) has improved the reasoning capabilities of Multimodal Large Language Models (MLLMs), yet existing frameworks rely on coarse, sequence-level reward signals that lack the fine-grained supervision over the visually-grounded steps within a multimodal reasoning chain. We investigate this gap through the lens of two token-level metrics: visual dependency (i.e. how much a token's prediction relies on the input image features) and predictive entropy. Our empirical analysis reveals two key findings: (1) correct reasoning chains exhibit a markedly sharper entropy reduction as visual grounding intensifies, compared to incorrect ones; (2) pivotal tokens, those whose misprediction triggers reasoning collapse, are statistical outliers in the joint distribution of visual dependency and predictive entropy derived from correct chains. Motivated by these findings, we propose token-level perception-grounded advantage estimation (TPAE), which estimates token-level advantages by measuring each token's statistical consistency with the vision-entropy patterns of correct rollouts. TPAE leverages this granular score to modulate the sequence-level advantage, producing a fine-grained supervision signal that can be integrated into various RLVR frameworks. Extensive experiments on seven benchmarks show that TPAE consistently outperforms leading strong baselines, yielding more stable and efficient optimization for multimodal reasoning. The code is publicly available at https://github.com/Zhihan72/TPAE.

Electromagnetic (EM)-reconfigurable antennas provide multiple candidate radiation patterns per element, thereby introducing an additional EM-domain degree of freedom. Integrating radiation-pattern reconfigurability, realized as EM-domain precoding, with conventional hybrid analog-digital precoding yields tri-hybrid multiple-input multiple-output (MIMO) precoding, which can substantially improve the spectral efficiency of wideband multi-user MIMO orthogonal frequency-division multiplexing (OFDM) systems. However, the joint design of EM, analog, and digital precoding remains challenging. To address this challenge, we propose a tri-hybrid precoding network (Tri-PNet) based on Conformer, an emerging neural architecture that combines the local modeling strength of convolutional neural networks with the global dependency modeling of Transformers. Furthermore, two representative radiation-pattern modes, i.e., the non-regular mode and the 3rd Generation Partnership Project (3GPP) Technical Report (TR) 38.901 mode, are investigated. Tri-PNet is trained in an unsupervised manner to jointly learn EM, analog, and digital precoding by maximizing the average sum spectral efficiency. Its radiation-pattern selection network (RPSNet) employs a Conformer encoder to capture both local and global frequency-domain correlations, whereas its hybrid analog-digital precoding network (HPNet) combines cross-attention and dual-path processing with singular-value-decomposition (SVD) and zero-forcing (ZF) priors. Simulation results under both radiation-pattern modes demonstrate that Tri-PNet outperforms random EM precoding and conventional hybrid MIMO without EM precoding, approaches the greedy EM precoding search scheme with substantially lower online complexity, and remains robust to imperfect channel state information (CSI).

Persistent spatial memory enables embodied agents to navigate familiar environments across repeated visits. However, targets may move while unobserved, including during navigation, making remembered locations unreliable by the time an agent arrives. Despite advances in memory retrieval and state prediction, accounting for continued hidden world evolution and revising beliefs under limited visibility remain challenging. We study Evolving-World Navigation, where agents infer target locations from intermittent observations, predict their states at inspection time, and revise beliefs using visual evidence. We propose EvolvingNav, which constructs a time-indexed belief from timestamped 3D object histories through a structured persistence-relocation model. The belief distinguishes persistence at the last observed location from relocation to alternative locations and retains probability mass outside the known candidate set. An event-driven filter propagates the current belief as time elapses, forecasts target occupancy at candidate inspection times, and incorporates new RGB-D evidence. Negative observations downweight location hypotheses according to calibrated, visibility-conditioned detection probabilities, while evidence tracking prevents repeated use of the same observations. A frozen, zero-shot vision-language controller uses the updated belief to choose actions and replan. We further introduce EvoWorld-Bench, a benchmark grounded in human activity traces, comprising 54 scenes and 803,680 tasks with controlled changes before and during navigation. In simulation and real-robot experiments, EvolvingNav improves navigation success and search efficiency over the evaluated baselines. Paired experiments show the clearest gains under learnable temporal patterns, while ablations demonstrate the value of preserving uncertainty and incorporating visibility-aware evidence.

In this paper, we investigate the componentwise linear property of support-two monomial ideals. Our first main result shows that if $I$ is a support-two monomial ideal, then its underlying simple graph $G_I$ is co-chordal; equivalently, by Fröberg's theorem, $\sqrt{I}$ admits a linear resolution. This phenomenon is quite rare for general monomial ideals. In fact, there exist monomial ideals with linear resolutions whose radicals fail to have linear resolutions, even when the radical is the edge ideal of a graph. For any monomial ideal $I$, one always has $\mu(I)\geq \mu(\sqrt{I})$. In the literature, support-two monomial ideals satisfying $\mu(I)=\mu(\sqrt{I})$ are of special interest, as they include edge ideals of simple graphs, weighted oriented graphs, edge-weighted graphs, and vertex-weighted graphs. We refer to such ideals as minimal support-two monomial ideals. We explicitly characterize all minimal support-two monomial ideals, as well as their powers, that admit linear resolutions. Next, we classify the linearity of non-minimal ones. Consequently, we obtain a complete classification of linear support-two monomial ideals, which shows that the property of being linear does not depend on the characteristics of the base field for support-two monomial ideals.

Score-based variational inference (VI) provides an alternative to Kullback–Leibler (KL)-based VI by minimizing the Fisher divergence between the variational distribution and the target. A prior score-VI approach formulates this optimization as an eigenvalue problem, with the variational distribution constructed from low-energy eigenstates. However, this eigenvalue-based formulation faces two high-dimensional obstacles: an intractably large parameter count due to exponential scaling and non-uniqueness of individual eigenvectors in degenerate or nearly degenerate low-energy subspaces. We propose QuanVI, a scalable quantum-inspired algorithm that combines a mixed-state density-operator formulation with a quantum tensor network (QTN) parameterization using the matrix product operator (MPO) structure. In degenerate low-energy subspaces, the density-operator formulation represents the subspace by its maximally mixed state rather than relying on a non-unique individual eigenvector, while the QTN parameterization compresses the density operator to avoid exponential parameter growth. Experiments and ablations show that QuanVI agrees with exact solutions in low dimensions and scales to high-dimensional synthetic and Bayesian posterior-approximation benchmarks, including challenging non-Gaussian targets.

In a recent study, Ekici and Yenmez (2026) characterize top trading cycles (TTC) on the unrestricted domain by strategy-proofness, individual rationality, and pair rationality, uncovering a correspondence between the axiomatic foundations of TTC and deferred acceptance, matching theory's two canonical rules. We show that their characterization fails on the single-peaked domain, but it holds on the single-dipped domain even without strategy-proofness. They also show that pair rationality can be replaced in their characterization by two weaker conditions concerning agents' first and second choices. In contrast, this replacement fails on the single-dipped domain, even when strategy-proofness is imposed.

Speaker diarization systems based on speaker embeddings and neural diarization exploit complementary forms of speaker information, but their intermediate representations are not directly compatible. We introduce Training-Free Affinity Fusion (TFAF), which integrates the speaker structure inferred by a neural diarizer into an embedding-based diarization system. The neural speaker partition is used to condition local speaker representations, from which we construct a continuous affinity matrix and combine it with the embedding-based acoustic affinity before a single global clustering step. The method requires no additional training, shared embedding space, speaker-label alignment, or hard transfer of the neural diarizer's speaker count. Experiments on AMI and CALLHOME show consistent DER improvements over both constituent systems; on AMI, fusion also improves speaker-attributed transcription. Ablations show that the neural speaker partition accounts for most of the gain, while retaining the continuous embedding-based affinities provides additional benefit over hard partition fusion.

FeTe has long been regarded as a nonsuperconducting antiferromagnetic metal with trivial band topology, but recent advances in stoichiometry control have begun to challenge this picture. Here we use higher-order epitaxy on zinc-blende MnTe to stabilize near-stoichiometric FeTe with strongly suppressed interstitial Fe and a superconducting transition onset near 13 K. Angle-resolved photoemission spectroscopy reveals markedly enhanced quasiparticle coherence, well-defined Fe-derived hole bands, and a nearly two-dimensional Dirac-cone-like state near the Fermi level. First-principles calculations identify an inversion between odd- and even-parity bands, yielding nontrivial $Z_2$ topology and a Dirac surface state consistent with experiment. These results elucidate the intrinsic electronic structure of stoichiometric superconducting FeTe and provide evidence for nontrivial band topology, positioning FeTe/MnTe as a promising platform for exploring topological superconductivity.

We study equivalences between quotient categories of module categories associated with Gabriel topologies. We establish necessary and sufficient conditions for a Morita context to induce an equivalence between quotient categories and characterize such equivalences in terms of Morita contexts after passing to rings of quotients. Motivated by Artin–Zhang's noncommutative Serre theorem, we introduce a notion of ampleness relative to a Gabriel topology and use it to characterize quotient categories of finitely generated modules under suitable noetherian hypotheses. As an application, we establish a sufficient condition under which the noncommutative Auslander theorem holds for actions of finite-dimensional Hopf algebras on AS-regular algebras without assuming that the Hopf algebras are semisimple.

For a map $F:\Omega\subset\mathbb{R}^n\to\mathbb{R}^m$, let $\Theta(F)$ be the $n$-form on $\Omega\times\mathbb{R}^m$ introduced by Tsai and Wang [arXiv:2604.04336]. If $\lambda_1,\ldots,\lambda_{r}$ are the nonzero singular values of $d F$ at any fixed point, we prove that $$ \Theta(F) \text{ has comass one if and only if }~ \mathcal{S}(\lambda):=\sum_{i=1}^{r}\frac{\lambda_i^2}{1+\lambda_i^2}\leq1. $$ Note that if $\text{rank} d F\le1$, $\mathcal{S}$ is always less than $1$. We also prove a dichotomy that if $\mathcal{S}\leq 1$, then either $\mathcal{S} < 1$ or $\mathcal{S}\equiv1$. When $\mathcal{S}<1$, the graph tangent plane is the only calibrated plane, generalizing the corresponding result for hypersurfaces. When $\mathcal{S}\equiv1$, we prove that $F$ is affine or of rank $2$.

We show that heavy dark baryons in a confining $SU(N)$ sector with a moderately large number of colors, $N\sim10$–$100$, can generate self-interactions spanning the full range of astrophysical velocities. Their chromo-electric polarizability induces van der Waals interactions that yield $\sigma/m_{\rm DM}\sim30$-$100~\mathrm{cm^2/g}$ at dwarf-galaxy velocities, decrease to the values required by galaxy clusters, and rise to $100$-$1000~\mathrm{cm^2/g}$ at velocities of a few $\mathrm{km/s}$, as suggested by the second perturber of JVAS B1938+666. Remarkably, for a given ratio $m_Q/\Lambda_d$, these observations can be used to infer the value of $N$. The resulting self-interaction phenomenology favors GeV-scale dark baryons, independently of the relic-abundance requirement. At these masses, a dark baryon asymmetry comparable to the visible one can account for the observed dark-matter abundance.

Video object removal presents a uniquely difficult editing challenge. Because a removal prompt specifies only what to erase rather than what to generate, the model must infer and reconstruct a highly specific occluded background entirely from the surrounding context. Existing training-free methods struggle with this because their editing mechanisms act primarily as localized erasers. They fail to actively synthesize the missing background details and often leave behind ghosting artifacts. To solve this, we propose TripleFlow, a training-free framework that tightly couples erasure and generation. It coordinates a source flow, a residual flow, and a synthesis flow throughout the entire process. By reusing a single target prediction, the residual flow isolates and suppresses the object, while the synthesis flow independently reconstructs the occluded background. Crucially, TripleFlow injects this newly synthesized background back into the editing trajectory at every step. This continuous feedback loop ensures that the generated structures actively guide the removal process, achieving seamless completion that is spatiotemporally consistent with the unedited scene. Extensive evaluations across five challenging benchmarks demonstrate that TripleFlow establishes a new state-of-the-art, significantly outperforming existing baselines in both reconstruction fidelity and temporal consistency.

Multi-unmanned aerial vehicle (UAV) networks in urban low-altitude environments couple UAV mobility, wireless access, and optical backhaul resources. Existing path-planning methods optimize flight distance or wireless signal quality, but can still concentrate traffic on shared optical backhaul links. We present Spatio-Temporal Wireless-Optical (STWO) planner, a backhaul-aware path-planning algorithm that jointly considers flight distance, wireless link quality, and time-varying optical-link offered-load ratio. STWO updates backhaul occupancy during sequential multi-UAV planning, allowing later UAVs to avoid congested optical paths while maintaining wireless connectivity. Experiments show that STWO reduces peak optical-link offered-load ratio by up to 56.4% and congestion ratio by up to 72.6% under dense UAV deployment, demonstrating the importance of wireless-optical awareness for reliable multi-UAV transmission.

When driven by a large enough electric field or current, the insulating state of vanadium dioxide ($VO_2$) typically collapses abruptly into the metallic state (the insulator-metal transition, IMT) following a brief incubation delay. These incubation delays, typically lasting microseconds or less, have been analyzed predominantly through macroscopic electro-thermal modeling, providing little statistical insight into the transition dynamics. Here, we investigate the electrically driven IMT in a two-dimensional $VO_2$ lattice and demonstrate that the material behaves as a complex, interdependent network despite being a single structural entity. Because metallic domains dissipate significantly more Joule heat than insulating regions, each switching event raises the Joule heating of the network, resulting in network dependency and triggering a cascading chain reaction. Combining time-resolved resistance measurements with a Joule-coupled resistor-network model and interdependent-network theory, we observe an abrupt (first-order) transition preceded by an anomalously long incubation plateau lasting thousands of seconds. During this metastable plateau, an effective branching factor of the switching activity approaches unity at criticality, and the plateau lifetime diverges with exponent $\zeta=1/2$; the order parameter scales with exponent $\beta=1/2$. These signatures coincide with those predicted for interdependent networks, indicating that dissipative Joule coupling alone can drive mixed-order catastrophic cascades in a single-network correlated-oxide system.

Deep research tasks require agents to navigate large knowledge spaces, synthesize evidence across many sources, and adapt their plans as findings emerge. Directed acyclic graph (DAG)-based multi-agent systems suit this setting because they support parallel execution and isolate each sub-task within a focused dependency context. Yet existing DAG-based agents instantiate a task-level plan before execution and repair the graph only after failures or missing evidence are observed. This Plan-then-Patch strategy is brittle for deep research: the system commits most strongly when its evidence is weakest, and later revisions waste computation on branches that should not have been planned. We propose DAGent, a DAG-based multi-agent framework with Evaluate-then-Grow incremental planning: an Orchestrator grows the task graph one batch at a time, conditioning each expansion on confidence and uncertainty signals from completed nodes. A hierarchical context layer propagates compact QueryDocs by default while preserving full execution traces for on-demand recall. The recorded DAG topology admits structural RL signals that outcome-only recipes cannot define; DAGRPO, a GRPO adaptation, injects topology-conditioned credit on Executor rollouts and a structural compliance regularization on Orchestrator plans. Across BrowseComp-Plus, GAIA, and xbench-DeepSearch, DAGent surpasses the strongest open-source baseline by 5.3 / 5.8 / 2.0 points at the Qwen3-235B-A22B scale, and the lead replicates across four open-source backbones and extends to GPT-5 at 327K context. At the Qwen3-8B scale, DAGRPO improves over a same-budget outcome-only GRPO baseline by 3.0 average Pass@1 points. A same-architecture comparison shows that evidence-conditioned planning reaches higher accuracy at lower per-task token, tool-call, and step footprints than its Plan-then-Patch counterpart. Code: https://github.com/hanwenliu6825/DAGent

Planning complex missions in unknown environments requires robots to reason simultaneously about what they should do and what they still need to discover. Existing approaches for solving LTLf missions typically assume a known environment, or separate the exploration of the environment from the execution of the mission, while semantic exploration methods look for one target at a time and ignore the mission being executed. To fill this gap, our main contribution is an adaptive high-level planning method that interleaves a task-driven semantic search with the execution of the mission, advancing both in a non-myopic manner. Our method leverages two representations built online, a metric-semantic scene graph, built with a Vision Language Model (VLM), that provides the evidence needed to locate the objects the mission refers to, and the deterministic finite automaton (DFA) encoding the mission, that indicates which of them matter at each mission state. At every planning stage, our planner selects the waypoints that are most valuable for both the semantic search and the advancement of the mission, valuing them over the remaining mission stages in order to avoid blocking states. The selected waypoints are then ordered in a single high-level plan, which is recomputed as new information arrives. In photorealistic indoor environments over five mission types, our method completes more missions than the compared approaches while having to cover less of the environment, and it does so with shorter paths and complying with the restrictions imposed by the mission.

We report the possible detection of a Dark Galaxy Candidate (DGC; AGC 322753), via the cross-matching of the Arecibo Legacy Fast ALFA (ALFALFA) and the FAST All Sky HI (FASHI) HI blind surveys. It is detected by both surveys at the same recessional velocity, with a comparable HI mass of $2\times10^{9}$ M$_\odot$, and with similar HI spectral features. The ALFALFA and FASHI spectra show a double-peaked profile, compatible with a disc-like ordered rotation, and an HI linewidth at 50 per cent of the peak ($W_{50}$) of $\sim$85 km s$^{-1}$. The common intersection of the two beams allows the source position to be constrained with much greater precision than in either survey individually. Being detected by both radio telescopes, a Radio Frequency Interference (RFI) origin is excluded. No evidence of an optical counterpart was found down to $\sim$27.5 mag arcsec$^{-2}$ in the DESI Legacy Imaging Surveys, which leads to a stellar mass limit of $M_\star \lesssim 5\times10^{7}$ M$_\odot$. Given this stellar mass limit, AGC 322753 has $M_{\rm HI}/M_\star \gtrsim 40$. The HI mass is higher than expected for totally starless haloes; this could imply that AGC 322753 is not completely 'dark' but could host a luminous counterpart below the current optical detection limit.

Does memory-dependent control need nonlinear recurrent dynamics? We study simulated air-hockey defence under temporary loss of puck tracking. A DreamerV3 teacher outperforms a memoryless policy under tracking loss, while resetting the teacher's recurrent state sharply reduces performance, which demonstrates that the task requires memory. We distil this teacher into compact recurrent policies with a 64 dimensional state, with a combination of a diagonal linear recurrence and an optional rank-$k$ nonlinear innovation while retaining nonlinear observation encoders and action heads. Across five matched seeds, the purely linear recurrent model ($k=0$) matches both the GRU baseline and the teacher throughout the tested range of tracking loss. Increasing nonlinear innovation rank providing no measured benefits. This result is obtained on a fresh test split, which will be only opened after all models and analyses are frozen. The linear model requires fewer recurrent parameters and less computation than GRU, but performs comparably. These results suggest that, for this memory dependent control task, nonlinear representation learning around a simple linear memory mechanism can be sufficient, and that nonlinear recurrent dynamics are not necessarily required. These conclusions are limited to the simulated task, teacher, state dimension, and blackout horizon considered here, and to policies whose observation encoder and action head remain nonlinear.

Omni-modal large language models deployed in real-world environments encounter external noise that can interfere with their perception and understanding of multimodal inputs. We study their robustness in audio-visual understanding, focusing on question answering under environmental noise and competing speech. The challenge is to resist acoustic interference while preserving useful audio evidence. On-policy distillation provides dense teacher feedback on student-generated responses, but uniform token weighting does not explicitly prioritize positions affected by acoustic interference. We introduce OP-CAD (On-Policy Clean-Audio Distillation), a curriculum-based privileged self-distillation framework for robust audio-visual understanding. Training progresses from mild to severe environmental noise and competing speech, with selective token-level supervision at each stage. The student generates responses from corrupted audio-visual input, while a frozen teacher uses clean audio and the verified answer to supervise the same response prefixes. To allocate this supervision, OP-CAD compares teacher predictions under clean, corrupted, and visual-only contexts without revealing the answer. These matched comparisons measure sensitivity to audio removal and corruption; a bounded weighting rule emphasizes positions identified by either signal while retaining supervision throughout the response. OP-CAD outperforms the compared methods across all evaluated noise conditions. Paired analyses further show improved preservation of clean-correct answers under strong interference, with no observed aggregate clean-accuracy penalty. These results demonstrate the value of directing clean-teacher supervision toward acoustically sensitive predictions for robust audio-visual reasoning.

Textual skills enable large language model (LLM) based agents to accumulate reusable procedural knowledge without updating model parameters. Yet existing skill evolution remains largely confined to the text space: an optimizer must diagnose success and failure patterns, and revise skills solely from long execution trajectories and sparse task outcomes. This text-only paradigm leaves the agent's internal representations, which contain rich records of its evolving execution state, outside the skill optimization loop. We ask whether an agent can improve its external textual skills by reflecting on its own internal representations. We introduce Rep2Skill, a representation-guided framework for self-evolution on agent skills. Specifically, upon the collected agent rollouts, Rep2Skill models their internal model representation trajectories to localize turns that deviate from successful execution dynamics, and it further interprets these signals alongside the execution contexts as actionable textual feedback for targeted skill revision. Experiments on two agent environments with two open-source LLMs show that Rep2Skill consistently outperforms text-only approaches in the self-evolution setting, where the same LLM serves as both executor and optimizer without a stronger external model. This establishes a promising direction moving agent self-improvement beyond text-only reflection.