Robotics
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Monday 9 November
Dexterous hands enable versatile manipulation, but policies trained in simulation often degrade on hardware owing to non-idealities, and the high-dimensional action space makes real-world adaptation costly. Piano playing, which requires precise and coordinated multifinger motion, offers a representative testbed. We propose GAP (Generalized Adaptation Prior for Dexterous Piano Playing), a sample-efficient adaptation framework that learns how to adapt, rather than how to act, during simulation pretraining and transfers this adaptation prior to hardware. Between repeated executions, GAP uses feedback to adjust the actions of the next one: a gradient guidance network learns from paired executions how adjusting actions affects playing performance, and its gradients guide these adjustments. In simulation, GAP yields approximately 2.4 times the F1 scores improvement of the existing fast-adaptation baseline for dexterous piano playing under the same interaction budget. The learned prior also generalizes to musical pieces held out from pretraining. On a physical dexterous hand, the same simulation-pretrained prior improves F1 scores by 3.6%-23% across three pieces of varying difficulty after approximately 7 min of adaptation each. These results suggest a practical route to reduce the real-world interaction needed to deploy dexterous hands.
Differential Geometry
Almost complex structures on triple blow-ups
We show that for certain smooth 4-manifolds, almost complex structures on its triple blow-up induce all its spin$^c$ structures with a non-negative-dimensional Seiberg-Witten moduli space. Since this includes all basic classes, our result provides an approximation to the simple type conjecture. Moreover, we expand a result by Dieter Kotschick showing that every finitely presented group is the fundamental group of an almost complex 4-manifold. We show that it can furthermore be chosen to have every even integer as the virtual dimension of the Seiberg-Witten moduli space of a spin$^c$ structure.
Read the full abstract
Artificial Intelligence
SkillDuet: Co-Evolving Models and Skills through Response Feedback
During reinforcement learning, experience can be both internalized in LLM parameters and retained externally as reusable textual skills. Parameter updates alter the model's behavior and the interaction experience from which new skills are extracted. These skills, in turn, guide the model's subsequent interactions and parameter updates. As the model's capabilities evolve, previously beneficial skills may become redundant or detrimental. We present SkillDuet, a framework that coordinates parameter updates and skill evolution through model response feedback. Skills extracted from on-policy trajectories guide parameter updates through OPD at every step. A response-based score, Strength, ranks candidate skills by the magnitude of the shift they induce in the current model's responses, and the highest-scoring skills are periodically consolidated into a persistent global skill set (Global). The updated model and Global jointly guide subsequent interactions, generating experience for further parameter and skill updates. With Qwen3-4B, SkillDuet achieves success rates of 98.57% on ALFWorld and 92.97% on WebShop, exceeding the matched SEED-API baseline by 2.86 and 5.47 percentage points, respectively. Ablations show that response-based skill selection and Global consolidation improve task performance. Further experiments show that the benefits of retained skills depend on both their content and their compatibility with the model. SkillDuet provides a concrete mechanism for model-skill co-evolution and skill-level recursive self-improvement.
Read the full abstract
Analysis of PDEs
Lie-Lions brackets of nonlocal vector fields
We study nonlocal vector fields on the space of probability measures with finite second moment. Using Lions derivatives, we define their Lie-Lions bracket, which adds to the spatial Lie bracket a correction due to the evolution of the law, and prove that its vanishing is equivalent to commutation of the global flows, under first-order regularity and linear growth assumptions. We then construct a Lie algebra of complete polynomial nonlocal fields. For this class, we establish commutator and second-order Baker-Campbell-Hausdorff formulas with cubic remainders in the Wasserstein distance $W_2$, controlled by the second moment and uniform in the number of particles. The proofs rely on finite-particle realizations, which preserve the bracket and jointly determine the nonlocal field. We also obtain uniform estimates for smooth bounded kernels and a Leibniz rule for the action of polynomial fields on polynomial observables. Examples from collective dynamics show that the vanishing of the spatial bracket at a fixed law is neither necessary nor sufficient for commutation. They concern symmetries, aggregation-dependent rotation, synchronization, the transmission of a turn along a hierarchy, effective three-body interactions and occupancy-dependent circulation.
Read the full abstract
Probability
Geometry and Stochastic Analysis of Flow and Diffusion Models: A Survey
Flow-based and diffusion-based models have become the principal means of learning and sampling high-dimensional probability distributions, from images and video to proteins, molecules and physical fields, and increasingly language. Each carries a simple law to the law of the data along a curve of probability measures, and behind the variety of their names lie three geometries, those of the state space on which the data live, of the space of probability measures through which the curve passes, and of the space of paths traced by the generating process. A single analytic idea runs through the subject. Every score identity used to train a diffusion model is an integration by parts, on the noise, on the data or on Wiener space, and Malliavin calculus extends this to nonlinear and infinite-dimensional diffusions. We develop these foundations with full definitions and follow them through normalising flows, flow matching and few-step maps, diffusion models and their convergence theory, conditioning, Schrödinger bridges and stochastic control, and their extension to manifolds, Lie groups, discrete and function spaces and jump processes. We close with statistical physics, renormalisation and, more speculatively, holography, in which the radial direction of anti-de Sitter space can play the role of flow time. A table of the main methods and a collection of open problems complete the survey.
Read the full abstract
Computers and Society
Perils and opportunities of AI for public climate services
The rapid integration of artificial intelligence (AI), particularly generative and machine-learning systems, in climate services development and provision is transforming how climate information is produced, accessed, and interpreted. AI-enabled climate services offer significant opportunities for process automation, optimisation, tailoring, and the translation of complex climate data into actionable knowledge for diverse users. However, the growing reliance on privately controlled algorithms, data infrastructures, and computing facilities raises critical concerns related to governance, transparency, equity and economic sustainability, among others. While the technical architecture of climate services increasingly relies on advanced machine learning and large-scale climate models, their legitimacy as public services may be weakened in the absence of arrangements that prioritise public value and long-lasting societal benefit over profit maximisation. The tension between the public mandate of climate services and the private nature of AI infrastructure challenges traditional notions of openness. Using a consensus-based elicitation method coupled with the PESTLE framework, we analyse the political, economic, social, technological, legal, and environmental risks and opportunities shaping the co-production value chain of AI-enabled climate services. We draw policy-relevant conclusions for data, model and infrastructure supporting the next generation of upstream and downstream climate services as critical applications for society. Our work presents potential and plausible challenges and benefits for a pan-European, decentralised public climate service built on sovereign AI infrastructure and open governance principles.
Read the full abstract
High Energy Physics (Phenomenology)
Theoretical constraints on beyond Standard Model extensions of the Higgs potential
A fundamental requirement for the consistency of quantum field theories is that their Higgs potential is lower-bounded, such that no direction in the $\Phi$ field space leads to $V(\Phi)\to \, -\infty$. This work presents a comprehensive analysis of the various theoretical constraints on potentials in extended Higgs sectors, highlighting their relative impact and restrictiveness. We illustrate our study on the potential parameters with two particular Higgs sector extensions: the Real Singlet Extension (RSE) and the Two-Higgs-Doublet Model (2HDM), separating $\mathbb{Z}_{2}$-conserving and $\mathbb{Z}_{2}$-violating cases. In order to do this, we derive and compare these conditions to other studies found in the literature. Additionally, we developed a publicly available code capable of testing positivity conditions in complex potentials, including particularly challenging cases such as the general $\mathbb{Z}_2$-violating 2HDM. Finally, we apply those restrictions to the physical parameters and deduce an upper and a lower bound on the masses of new heavy scalars.
Read the full abstract
Combinatorics
Permutons
Permutons are limit structures for large permutations. Intrinsically, a permuton is a familiar mathematical object: a probability measure on the unit square with uniform marginals. The compact space of permutons completes the set of permutations of all sizes under a natural topology; each permuton is not only a limit of permutations but also a generator of random permutations. Since their (relatively recent) discovery as a tool for understanding the behavior of large (possibly random) permutations, permutons have proved remarkably fruitful, with dozens of fascinating connections to combinatorics and probability. We review the fundamentals of permutons, sometimes providing new, shorter proofs, and give an account of some of their myriad applications. The survey concludes with a long list of open problems. An extensive – but not comprehensive – bibliography is appended.
Read the full abstract
Graphics
Real-Time Volumetric Light Transport Inference without a Path
We present a real-time method for reconstructing fully integrated pixel radiance in rendered volumetric media under direct and ambient environmental illumination, using compact, low-cost scene features and statistics. Unlike Monte Carlo denoising pipelines for volumetric rendering, which depend on computationally expensive path-traced radiance estimates, our approach directly infers the final light-transport solution from inexpensive primary-ray and environment sample evaluations. We show that these features form a near-sufficient representation of the scene, enabling accurate reconstruction at substantially lower computational cost. Using an attention-augmented image-to-image network, we demonstrate that complete pixel radiance can be inferred directly from these auxiliary features, without requiring the original radiance signal. Operating in image space also provides a more stable computational cost as the camera zooms into or out of the volume, with substantially less dependence on volume density than approaches based on explicit multiple-scattering simulation. We demonstrate real-time performance and high reconstruction accuracy in challenging volumetric cloud scenarios.
Read the full abstract
Robotics
SPARK: Simple Post-training for Adapting pRetrained Knowledge to Robot Control
Large-scale representation learning has produced public foundation models that provide strong general-purpose visual features. However, their pretraining objectives are not designed for robot representation learning, so the resulting embeddings tend to emphasize global semantics rather than the temporally sensitive representations required for robot control. Training robot models from scratch is also undesirable due to the limited scale and high collection cost of robotics data, as well as the substantial computational burden of pre-training. In this paper, we propose SPARK (Simple Post-training for Adapting pRetrained Knowledge), a post-training method that adapts foundation models to robot control. SPARK is built on two principles: temporal abstraction, which encourages the encoder to derive compact features that preserve information essential for understanding temporal changes and action-relevant scene structure, and knowledge preservation, which aligns patch-level representations with those of the original foundation model to retain useful pretrained semantics. These objectives yield compact visual state representations that remain semantically meaningful while preserving useful prior knowledge. Experiments across multiple robotics benchmarks show that SPARK consistently improves success rates and generalization over vanilla foundation models, and further demonstrate that these gains transfer to real-world robot manipulation.
Read the full abstract
Differential Geometry
Branched real projective structures on surfaces, cyclic Higgs Bundles and Anosov representations
We use the theory of $3$-cyclic $\mathrm{SL}(3,\mathbb{R})$-Higgs bundles to study real branched projective structures on compact surfaces recently introduced by the authors. After fixing a Riemann surface structure on $S$, the Higgs bundle approach allows us, on the one hand, to construct branched structures with prescribed branching data and, on the other hand, to introduce new classes of such structures. The latter are directly related to singular hyperbolic affine spheres and carry a Blaschke metric with conical singularities. In the final part of the paper, using different methods, we prove that every Anosov surface group representation into $\mathrm{SL}(3,\mathbb{R})$ is either Hitchin or belongs to the Barbot component. As a by-product, we obtain that some of the representations associated with the branched structures constructed above are not Anosov.
Read the full abstract
Cryptography and Security
FALC: An Artifact-Centric Life Cycle Model and Research Agenda for AI Forensics
AI systems increasingly appear in criminal investigations as tools, targets, and sources of evidence, yet AI forensics lacks a systematic account of the artifacts they produce: investigators often do not know which traces to look for, and for many traces they do find, no established analysis technique exists. In this paper we first developed the Forensic AI Life Cycle (FALC) that unlike other life cycle models shifts the focus onto the forensic artifacts created during the life cycle of an AI system. Through FALC we derived a reference list of artifacts that might serve an investigation. Second, we collected examples for crimes with AI involvement throughout FALC and questions likely to be asked when investigating these crimes. Third, we propose a taxonomy of forensic fields of expertise with a focus on the current possibility to analyze AI forensic artifacts using established techniques. For investigators, the model provides a systematic basis for identifying which artifacts to look for when an AI system is relevant to a case. For researchers, the classification shows where existing forensic methods already suffice and where new techniques are most urgently needed.
Read the full abstract
Robotics
SAM-VTP: Video Memory and Measured Geometry for Continuous Traversability Segmentation
Traversability prediction is a core component of autonomous navigation, yet most vision-based methods operate frame-wise and produce temporally inconsistent predictions. We present SAM-VTP (Video Traversability Prediction), which fine-tunes SAM2's video memory pathway for dense, prompt-free traversability segmentation, conditioning each frame on a persistent memory of prior observations. Rather than inferring terrain geometry from appearance, we measure it: slope, roughness and height are computed from depth and fused into the image features. We also introduce an offline continuous supervision pipeline that generates dense per-pixel traversability targets in [0,1] from depth-derived geometry and semantic priors, requiring no robot deployment and no per-pixel annotation, capturing conditional feasibility that binary labels cannot represent. On TartanGround, a large-scale synthetic ground-robot dataset, SAM-VTP improves fuzzy Jaccard from 0.908 to 0.924 and binary IoU from 0.854 to 0.886 over SAM-TP, a frame-wise SAM2 baseline, against a seed-to-seed spread below 0.001, and cuts frame-to-frame flicker by 31%. In closed-loop trials at two field sites, driving the same planner and controller, the baseline's dropouts leave the planner without a feasible path 3.3x more often on a paved course and 6.9x on rough ground, while SAM-VTP runs at 21.4 FPS on the Orin Nano the robot carries.
Read the full abstract
Quantum Physics
A Universal Fault-Tolerant Quantum Computing Architecture with Globally Driven Atom Arrays
Scalable quantum computation requires reconciling programmability with simple, high-fidelity control. Here, we propose a universal fault-tolerant architecture using globally driven dual-species atom arrays. In a hybrid neutral-atom system, interspecies Rydberg interactions enable qubit addressability without locally addressed quantum gates. We construct a universal native instruction set, Global Rydberg Atom Instruction Language (GRAIL), and show that crosstalk errors can be suppressed to the $10^{-4}$ level. We compile surface-code error correction and magic state cultivation into GRAIL, and parallelize the compilation to suppress crosstalk. Under a hardware-inferred circuit-level noise model, we obtain an error-correction threshold of $1.43\%$ for the resulting quantum memory and a logical error rate below $10^{-9}$ for magic states cultivated at an experimentally accessible Rydberg Rabi frequency. By replacing local gate addressing with atom-mediated selectivity, this architecture removes the associated calibration requirements, offering a route to simpler control for large-scale fault-tolerant neutral-atom computation.
Read the full abstract
Machine Learning
Calibrating Document-Local Boundaries for Contextual Chunk Retrieval
Late Chunking encodes an entire document before pooling at chunk boundaries, preserving document context in local retrieval units. Chunk-level post-training improves context-dependent retrieval, but shared context can make semantically similar same-document negatives difficult to distinguish. We study the local decision boundary between a gold chunk and its hardest sibling. Starting from InSeNT, we analyze InSeNT's within-document gradients and show that the entropy of the conditional negative allocation decreases monotonically with the scaling factor at fixed logits. We propose a two-part framework for calibrating document-local chunk boundaries. The first component adapts the discriminative scale from within-document margins, while the second exposes chunk partitions to attention during document encoding. Experiments on ConTEB and NanoBEIR show that the resulting model improves contextual chunk retrieval across the evaluated encoder settings while retaining competitive passage-level retrieval quality; ablations further characterize the distinct roles of adaptive scaling and chunk-aware routing.
Read the full abstract
High Energy Physics (Theory)
Two-Loop $n$-Point Identical Helicity Amplitude for Gravity Minimally Coupled to a Chiral Fermion
Motivated by recent developments in the two-loop identical-helicity sector of gauge theory when the one-loop self-dual integrability anomaly of Bardeen and Cangemi, or the equivalent twistor-space version of Costello and Paquette, vanishes, we compute the two-loop four-point identical-helicity amplitude in gravity coupled to a single chiral fermion, for which the corresponding self-dual anomaly also vanishes. We show that the infrared-subtracted amplitude is purely rational and proportional to the identical-helicity tree amplitude of pure gravity with a single insertion of the operator $R^3$. Using this connection, we obtain a closed-form expression for the two-loop $n$-point identical-helicity amplitude in gravity minimally coupled to a fermion. We do so by first obtaining the corresponding identical-helicity gauge-theory tree amplitude with a single $F^3$ insertion using a chiral bootstrap. We then convert these $n$-point gauge theory amplitudes using Kawai-Lewellen-Tye to obtain gravity amplitudes with an $R^3$ insertion, which are necessarily equal to the desired two-loop identical helicity amplitudes of gravity coupled to a chiral fermion. As a nontrivial check, we explicitly verify crossing symmetry and the expected soft limits of our expressions through 13 points.
Read the full abstract
Artificial Intelligence
From Fluent Text to Ontology-Compatible Definitions: A Quality-Control Agentic Workflow
Biomedical ontologies are structured knowledge systems that must evolve continuously to represent newly emerging knowledge, but their manual maintenance remains time-consuming and laborintensive. Textual definitions are especially costly because they must be readable to humans and compatible with controlled entities, relations, and logical constraints. Large language models (LLMs) offer a promising way to assist ontology engineering tasks, including definition generation. However, textual plausibility does not guarantee knowledge-system compatibility. A generated definition may be fluent and close to a reference definition, yet still fail to ground its implied entities or violate ontology-level constraints. Directly inserting such definitions into ontology libraries may therefore introduce entity-level hallucinations and logical inconsistencies. To address this problem, we propose two structured-knowledge incompatibility metrics, Entity Error Rate (EER) and Logical Inconsistency Rate (LIR), to quantify entity grounding and logical consistency for LLM-generated ontology definitions. We further develop an agentic quality-control (QC) workflow that treats definition generation as an iterative process rather than a direct decoding step: it generates definitions, extracts axioms, checks entity existence and logical consistency, and uses the resulting diagnostics to drive repair loops. Comprehensive experiments on Disease Ontology, Uberon Ontology, and Uberon-ConstraintStress demonstrate that EER and LIR complement conventional reference-based metrics by evaluating whether generated content can be grounded and checked inside the target ontology. The workflow effectively decreases EER while maintaining strong conventional text-generation performance. LIR further exposes sparse ontology-level logical failures that are invisible to text similarity metrics, although their repair remains limited.
Read the full abstract
Populations and Evolution
Introduction to genetic simulation for animal breeding and genetics
This article introduces genetic simulation for animal breeding and genetics. We will discuss some of the reasons to perform simulations in science, terms used to describe different types of genetic simulations, examples of common simulation approaches and tools that implement them, and how to get started with simulation in your research. The three simulation programs discussed are the population genetic coalescent simulator msprime, the evolutionary forward-in-time simulator SLiM and the quantitative genetic simulator AlphaSimR, with an emphasis on the latter as it is the most important for animal breeding and genetics. An example is also given about how to write simple population genetic simulations from scratch in a scripting language like R. Small code examples are provided.
Read the full abstract
General Relativity and Quantum Cosmology
Next-to-next-to-leading gravitational radiation-reaction corrections to quasi-elliptic and quasi-hyperbolic compact binary orbits
We apply the Lagrange method of variation of constants to compute the radiation-reaction (RR) corrections to the quasi-elliptic and quasi-hyperbolic motion of compact binary systems at 4.5PN order, i.e., at next-to-next-to-leading order in the RR. We rely on the recent determination of the 4.5PN RR acceleration, in a general frame and in the center-of-mass frame, in harmonic coordinates. The non-local effects, due to the tails at 4PN order and to the gravitational-wave recoil at 4.5PN order, are carefully evaluated. Although the calculation is implemented differently in the quasi-elliptic and quasi-hyperbolic cases (different perturbation equations and choices of orbital parameters), we verify by analytic continuation that the results for the local effects are equivalent in the two cases.
Read the full abstract
Artificial Intelligence
CRAFT-AI: Capability and Requirement Alignment for Flexible Manufacturing through Agentic AI
One of the fundamental needs for reaching resilient production infrastructure is flexible manufacturing, which requires translating changing production requirements into implementable system configurations. This paper introduces CRAFT-AI, an agentic AI approach combining natural language (NL) requirement interpretation, Asset Administration Shell (AAS)-based capability descriptions, and reusable engineering knowledge from existing manufacturing configurations. A tool-using AI agent matches requirements with available capabilities and reference implementations to generate manufacturing scenes and corresponding AAS instances. The approach is demonstrated using a Factory I/O prototype with twelve reference scenes and five illustrative manufacturing tasks of varying complexity, providing an initial feasibility assessment of the proposed AAS-enabled, system-level generative engineering approach. For the four valid user task descriptions, proper requirements, correct references to the relevant engineering knowledge, correct components and lastly a valid Factory I/O scene are generated. For the task with missing engineering knowledge and a wrong sensor selection, the user is informed about which requirements of the scene generation lack domain knowledge and also which sensor is not compatible with the selected raw material by the user with link to the related documentation. After fixing the task description, the correct scene is seamlessly generated. CRAFT-AI highlights the importance of safeguards for risk perception and communication in foundation model-driven solutions.
Read the full abstract
Soft Condensed Matter
Topology, geometry and fabrication of volumetric plain weaves
Plain weave is one of the simplest woven structures, formed by two sets of yarns that cross at right angles in an alternating over-under pattern. While such woven fabrics are usually confined to a 2D surface, here we propose a practical approach to fabricate volumetric plain weaves, entanglements featuring three mutually orthogonal families of yarns threaded through a cubic lattice. We develop an analytic framework that converts the crossing pattern topology into continuous yarn centreline geometries, which we then leverage to fabricate volumetric weave specimens. Each crossing is encoded by a matrix that defines the offsets of the three yarns at a vertex. These are formulated to preserve the plain weave topology in every plane, where each pair of yarn families interlaces as in a planar plain weave. Eight configurations in four opposite pairs exist, yet all are related by rotations and translations, so that the fully alternating volumetric plain weave has a single topology. We adopt sinusoidal yarn trajectories and show that the crossing matrix defines their phases. We select two solutions: in-phase planar zigzags and out-of-phase circular helices. Both are members of a family of elliptical helices that also admits a free choice of handedness for each yarn. We then derive the conditions under which yarns intersect between vertices. In-phase yarns never intersect, whereas helical yarns of equal handedness intersect at a critical offset of $\sqrt{2}\lambda/16$, with $\lambda$ the wavelength of the cubic lattice. No assignment of handedness avoids these intersections. Finally, we fabricate volumetric plain weaves by 3D printing and manual assembly from cotton cord. The framework provides a direct basis for generating weave geometries for simulation and fabrication.
Read the full abstract
Machine Learning
TSFFormer: A Frequency-Domain Spike Transformer for Green Human Activity Recognition
Wearable Inertial Measurement Units (IMUs) enable pervasive Human Activity Recognition (HAR), but traditional Deep Neural Networks (DNNs) suffer from an "arithmetic-latency deadlock" due to dense floating-point MAC operations. While Spiking Neural Networks (SNNs) offer ultra-low power, existing ones face spatial blind spots, rigid neurons, and gradient degradation. To address these, we propose TSFFormer, a purely spiking Fourier Transformer for Green HAR. It introduces two key innovations: (1) a MultiHeadFourier attention that performs frequency-domain global mixing, reducing complexity from \(O(V^2)\) to \(O(V\log V)\), eliminating spatial blind spots, and saving \(\sim\)50% GPU memory; and (2) a dual-compartment Temporal Spiking LIF (TSLIFNode) that utilizes dynamic thresholds to adapt to non-stationary rhythms and mitigate gradient degradation. Extensive experiments on 8 public IMU benchmarks demonstrate that TSFFormer surpasses state-of-the-art SNNs and even the classic DeepConvLSTM, achieving a \(\sim\)3\(\times\) faster training speed with an energy footprint of 0.1 pJ. Ablation studies confirm the synergistic effect of our modules, providing a highly accurate, ultra-low-power event-driven paradigm for wearable sensing.
Read the full abstract
Optimization and Control
Operational Interpretation of Controllability Allocation Scores via Randomized Interventions
Controllability allocation scores (CASs) assign weights to virtual input channels without specifying how interventions should be implemented. This paper provides an operational interpretation through randomized finite-duration interventions with prescribed, possibly candidate-dependent waveforms. Selection probabilities are fixed before a mean-output target is specified, while waveform amplitudes are optimized after the target is revealed. We derive the unique minimum expected-energy amplitude policy and formulate volumetric and weighted average-energy allocation criteria. We also establish a finite-trial mean-output error bound and show that waveform-based weighted AE-CAS minimizes an upper bound on the target-averaged mean-square error. We establish uniform convergence of normalized waveform-response Gramians and show that accumulation points of optimal allocations solve the limiting problem. With uniform timing, equal positive waveform efficiencies recover the original CAS in the short-duration limit, whereas unequal efficiencies yield waveform-weighted limits. Under the former conditions, the limiting error bound is proportional to the original weighted AE-CAS objective. For common waveforms, we further establish an exact finite-duration correspondence with the original full-state V-CAS under uniform timing.
Read the full abstract
Machine Learning
Zero-Search Reinforcement Learning for Slay the Spire 2 with Failure-Driven Restarts
Slay the Spire 2 (STS2) is a deck-building roguelike in which a winning run at the highest difficulty takes several hundred decisions, and most losses happen in late boss fights that a learner seldom reaches. We train agents that choose every action with one forward pass of a 6.6M-parameter entity transformer, without lookahead. Training and evaluation run the game's own compiled assembly (v0.107.1) in a headless .NET host. The host replays a run exactly from its seed and actions, so training episodes can restart from states the learner has visited, and a failure-driven curriculum allocates these restarts in proportion to where complete runs are lost. On 12,800 held-out games at Ascension 10 (A10), the released policies win 8.05% averaged over the five characters (Ironclad 15.6%, Regent 12.5%), and AlphaSpire's released networks used without search win 0.55% on a common panel. Longer A10 training raised Ironclad from 12.4% to 15.6% and Defect from 3.0% to 8.1% on these games. At Ascension 0 the released policies win 80.2%, level with the 80.3% that the tree-search agent Spire Mentor reports, while Mentor leads by 5 to 8 points from A4 upward. Run in our host on the same 2,000 A10 games, Mentor's search wins 8.8% against 8.2% for our policies, which need 21-27 ms per decision on one CPU core. GPT-6 Astra at maximum reasoning won none of 11 A10 games through our text interface. In a controlled study of 18 training runs from six random initializations, failure-driven allocation exceeds uniform restart allocation on each of three Regent initializations, by 3.6 percentage points on average, while the Silent effect is small and changes sign. Code, trained policies, an in-game mod and the records behind every table are at https://github.com/HalfVulpes/sts2ai.
Read the full abstract
Computer Vision
Compass 1: Towards Open-World and Persistent Embodied Navigation from Monocular RGB Videos
Embodied navigation requires agents to achieve persistent spatial awareness and long-horizon execution stability in open-world environments. However, existing navigation architectures struggle to sustain both capabilities, remaining effective only over a handful of recent frames. We present Compass 1, the first embodied navigation foundation model to unify language understanding and video generation while enabling memory backtracking through nearly 2mins of visual history. Without altering the underlying architecture, we introduce History Representation, which integrates sparse visual keyframes with an explicit History Representation Map to preserve long-term spatiotemporal context. To effectively leverage this history, we further introduce History Utilization, a structured VQA supervision protocol comprising three categories: task-stage grounding, exploration consistency, and spatiotemporal target retrieval, enabling map-grounded reasoning and stable persistent long-horizon navigation. To fuel this system, we establish a complete annotation pipeline for monocular RGB videos and open-source a massive dataset featuring over 500 hours of high-quality real-world interactions and over 2,000 hours of annotated web videos. Trained mainly on the distribution of 1m height camera, Compass 1 generalizes seamlessly across diverse embodiments and scenarios, ranging from indoor environments to pedestrian-dense streets, spanning wheeled robot at 0.35m, quadruped dog at 0.5m, humanoid at 0.85m, and aerial drone at 1.5m. Compass 1achieves a dominant overall success rate of 79.4% across open-world evaluations, compared to 17.8% for the strongest baseline. It demonstrates exceptional zero-shot execution exceeding 90% on foundational tasks, a 250% boost in exploration, and an unparalleled leap from 0% to 58% in challenging persistent history backtracking.
Read the full abstract
Nuclear Experiment
Proton Number Fluctuations in Au+Au and Ag+Ag Collisions at $\sqrt{s_{\mathrm{NN}}}=2.4\ \mathrm{GeV}$: A JAM Model Study
Proton number fluctuations in low-energy heavy-ion collisions are sensitive to nuclear interactions, baryon-number conservation, and fluctuations of the collision volume, all of which can complicate searches for critical phenomena. We investigate proton-number cumulants and factorial cumulants up to fourth order in Au+Au and Ag+Ag collisions at $\sqrt{s_{\mathrm{NN}}}=2.4\ \mathrm{GeV}$ using the JAM hadronic transport model in cascade and mean-field modes. We employ the Centrality-Independent Genuine Cumulant Analysis fRamework (CIGAR) to suppress the initial-volume fluctuations and compare its results with those obtained using conventional centrality-bin-width corrections. CIGAR closely reproduces the participant-number-based reference, whereas the multiplicity-based analysis exhibits substantial deviations. In both collision systems, the ratios $C_{2}/C_{1}$, $C_{3}/C_{1}$, and $C_{4}/C_{2}$ generally decrease toward more central collisions and with increasing rapidity acceptance, while the mean field modifies their magnitudes. A simplified treatment of light-nucleus formation through probabilistic proton removal shifts $C_{4}/C_{2}$ toward less negative values. These results establish a hadronic-transport reference without an explicit QCD critical contribution and highlight the importance of controlling centrality fluctuations, acceptance, and light-nucleus formation when interpreting experimental measurements at high baryon density.
Read the full abstract
Image and Video Processing
Next-Generation Segmentation in Radiotherapy: A Narrative Review from Atlas-Based Methods to Agentic Vision-Language Models
Precise segmentation of target volumes and organs at risk is an essential prerequisite for modern, high-precision radiation therapy. However, this task has traditionally been time-consuming and subject to high interobserver variability. Motivated by these limitations, more advanced and novel approaches to automating this task have been developed over the past two decades. This narrative review highlights representative milestones, with a focus on the most recent developments. These approaches include atlas-based label propagation, convolutional architectures, promptable foundation models, and, most recently, vision-language models (VLMs), which integrate clinical text with volumetric segmentation. Each methodological generation has addressed limitations of its predecessor while introducing new ones. A substantial gap remains between research progress and clinical deployment, as routine workflows continue to be dominated by atlas-based and narrow task-specific deep-learning tools, which remain strong baselines in geometric overlap metrics but are highly rigid in terms of application, versatility, portability, and interpretability. As contouring systems further grow in complexity, conventional geometric overlap metrics become insufficient, motivating evaluation paradigms that incorporate surface-based, dosimetric, and prompt-adherence assessment. Emerging agentic architectures that combine large language models, retrieval-augmented generation, and vision-language segmentation modules are leading towards more flexible, guideline-aware, interactive contouring workflows. These systems could automate flexible and interpretable knowledge-based contouring and actively support the clinical reasoning underlying target volume definition.
Read the full abstract
Numerical Analysis
A Computational Brainsphere Model for the Simulation of Alzheimer's Disease
Nowadays, there are different medical imaging techniques to collect data about the progression of Alzhei\-mer's disease in a patient's brain. These data describe different phenomena which are still not understood from a biological point of view. How can these data sets be combined in a mathematical model to simulate the evolution of such a neurodegenerative disease in a computer? In this snapshot, we present one possible approach to address this task with the help of graph theory and partial differential equations.
Read the full abstract
Software Engineering
SWE-Time: Benchmarking Temporal Repository Understanding
Repository-level code-understanding benchmarks have progressively expanded the spatial scope of reasoning, from local code to entire repositories, while largely treating each repository as a static snapshot. However, how models reason about repository evolution over time remains largely unexplored. We introduce SWE-Time, a benchmark for temporal repository understanding that makes repository evolution itself the reasoning target. SWE-Time evaluates two complementary temporal directions: retrospective understanding of repository evolution through historical regression analysis and prospective understanding of intended changes from pre-change repository states. To enable realistic yet reliable evaluation in these settings, we introduce the RealWorld-to-Verifiable Construction Framework, which transforms real-world developer queries and historical evidence into verifiable evaluation tasks. We evaluate 15 coding agents on 1,661 cases from 65 GitHub repositories. Within Claude Code, Claude Opus 4.8 achieves the highest SWE-Time Avg. of 61.9, leaving substantial headroom. Controlled analyses highlight the importance of historical code evidence and developer intent, while consistent reasoning across repository states remains challenging. These results highlight repository evolution as a distinct and challenging dimension of code understanding.
Read the full abstract
Signal Processing
Prospective Importance-based Cognitive Pilot Design for Semantic Sensing in ISAC Systems
Semantic sensing enables integrated sensing and communication (ISAC) systems to acquire environmental information relevant to downstream tasks. However, existing pilot design methods either rely on static pilot configurations or learn end-to-end pilot selection policies, without explicitly evaluating the semantic contribution of candidate pilot locations. In this work, we propose a cognitive semantic sensing framework for human activity recognition in orthogonal frequency division multiplexing (OFDM)-ISAC systems. Rather than directly learning a high-dimensional mapping from observations to pilot locations, we introduce a semantic importance weight (SIW) to measure the semantic contribution of each candidate pilot location. Leveraging the information bottleneck (IB) principle and the mutual information-minimum mean-square error (I-MMSE) relationship, SIW assigns high values to future time-frequency (TF) channel coefficients that are both difficult to infer from the preceding observations and valuable for downstream inference. To efficiently obtain the SIW, we further develop a conditional diffusion model that predicts the corresponding uncertainty through denoising covariance estimation. Experiments on a real-world Wi-Fi sensing dataset show that the proposed method consistently outperforms baselines across different signal-to-noise ratios (SNRs) and pilot budgets.