Search arXiv⌕ Search

arXiv subjects

Lai Wei

Publications and source records attributed to Lai Wei.

At least 19 recordsLinked to original sources
arXiv2026

HuG: Learning Humanoid Whole-Body Affordance-Aware Grasping with Active Perception

Acquiring objects in human environments is hard for humanoids: targets are partially occluded under egocentric viewpoints, and reaching them demands coordinated whole-body motion that keeps the target visible and reachable while the base stays stable. We present HuG, a framework for whole-body affordance-aware grasping with active perception that couples perception and action through a Perceive-Align-Grasp pipeline: belief-driven active perception resolves occlusion by treating viewpoint as an explicit decision variable; a point-cloud-conditioned affordance policy proposes whole-body grasps; and an affordance-aware alignment stage repositions the body to the most executable configuration. A key insight is that, on a floating base, kinematic validity is not enough: we curate training data with controller-in-the-loop dynamics simulation, rejecting 73% of kinematically valid grasps that would destabilize the deployment controller, so the policy learns kinodynamically feasible affordances. Trained entirely in simulation with zero real demonstrations, HuG attains 87.5% real-world success on a multi-level shelf task, generalizes to novel objects, and holds at least 73% success at every tested height. Beyond generalizing within its training distribution, the policy extrapolates spatially: at scale in simulation it succeeds at table heights well below any seen during training, indicating that kinodynamically filtered synthetic data grounds affordances in object geometry and whole-body feasibility rather than in memorized configurations.

cs.RO↗
arXiv2026

Zooming without Zooming v2: From Seeing Details to Connecting Evidence for Generalizable Multimodal Perception

Multimodal perception forms the foundation of multimodal intelligence. In particular, spatial perception and multimodal long-context understanding are two pivotal tasks that underpin embodied AI and multimodal agents. However, synthesizing targeted, high-quality training data for these tasks remains difficult and costly. Our key insight is that these tasks share an underlying capability, which we call connecting evidence: identifying relevant visual facts, binding them to the correct entities, recovering their relations, and composing them into an answer. Crucially, targeted training data are far easier to synthesize for this capability than for these tasks themselves. To strengthen this capability, we propose EvidenceConnect, an easy-to-implement data synthesis framework that uses zoomed views of related regions within a single image to generate questions requiring comparison, aggregation, or chaining of visual evidence. To ensure quality, a reference answer is retained only when independent VLMs agree on it, and region-ablation tests discard questions whose answer survives the removal of selected regions. We then train base VLMs on the resulting data with RLVR, obtaining Zooming without Zooming v2 (ZwZ2). Although the synthesized data are not tailored to any downstream benchmark, ZwZ2 improves fine-grained and spatial perception, with further gains on embodied spatial reasoning, multimodal long-context understanding, multimodal agent tasks, and general visual reasoning. These results establish EvidenceConnect as an effective framework for supervising evidence connection and suggest that strengthening this shared capability leads to more generalizable multimodal perception.

cs.CV↗
arXiv2026

TCSP3.0: Template-Based Crystal Structure Prediction with Coefficient-Preserving Element Substitution and Structural Diversity

Template-based crystal structure prediction (CSP) transfers a known crystal structure to a new composition by element substitution, so its accuracy depends on retrieving suitable templates, mapping elements and ranking the resulting candidates correctly. Auditing the original TCSP2.0 implementation on the T180 benchmark, we found that 469 of its 2,347 selected structures (20%) have the wrong composition, caused by element mappings that ignore stoichiometric coefficients and by sequential text substitution that cascades or corrupts element symbols. We propose TCSP3.0, which enforces coefficient-preserving mappings, substitutes species simultaneously on parsed structures and validates every candidate's composition and cell. It further adds periodic-number retrieval, soft oxidation-state evidence, radius-compatibility scoring, structural deduplication, and ORB-v3 machine-learning-potential relaxation and energy ranking of its candidates, as used by deep generative CSP baselines. TCSP3.0 achieves Top-1 and Top-20 match rates of 72.2% and 90.6% on T180, compared with 28.3% and 45.6% for TCSP2.0, and a Top-20 match rate of 74.0% on a new 800-target benchmark (T800). On T180 it also outperforms PNcsp+ and deep generative models CrystalFlow and MaskGXT (Top-1 72.2% versus at most 63.9%). Even without relaxation, TCSP3.0 reaches 67.8% at Top-1 and 90.6% at Top-20 in a median of 28 s per target on two CPU cores, about ten times faster than CrystalFlow. Paired ten-arm ablations on T180 and T800 show that most of the improvement comes from the correctness fixes; structural deduplication significantly improves the Top-5 match rate, embedding-based retrieval is required for coverage, and the remaining scoring features have small or null effects.

cond-mat.mtrl-sci↗
arXiv2026

Compositional Humanoid Manipulation through Isomorphic Teleoperation and Independent Skill Routing

Humanoid manipulation requires reusable arm skills, yet demonstrating every possible skill combination is impractical. We present Humanoid Semantic Mixture-of-Experts Diffusion Policy (HS-MoDP), which composes separately demonstrated arm skills through independent semantic routing over a shared expert pool. This design preserves arm-specific skill control while sharing action-generation parameters across arms. A semantic scheduler integrates the learned policy with an existing locomotion controller through alternating manipulation and repositioning stages, extending the effective workspace and enabling skill reuse across location changes. To acquire demonstrations of whole-body contact manipulation, we develop G1Exo, an isomorphic exoskeleton that provides direct joint-level control of arm configurations. Experiments on a Unitree G1 show stronger skill composition than the baselines. G1Exo enables more reliable contact teleoperation than end-effector VR-IK, and its demonstrations improve contact-skill learning across four policy architectures. For HS-MoDP, switching from VR-IK to G1Exo demonstrations raises full cabinet workflow success from 35% to 65%. All project hardware designs and source code will be released as open source after the review process.

cs.RO↗
arXiv2026

Deep Generative Crystal Structure Prediction: A Benchmark Study and a Controlled Test of Prototype Dependence

Deep generative models are widely reported to enable de novo crystal structure prediction (CSP), but their capability has not been measured consistently against template-based methods. We evaluate 12 representative generative CSP models, spanning latent-variable, diffusion, flow-matching, autoregressive, and manifold random-walk architectures, against TCSP 2.0 on 180 test structures and a leakage-controlled subset of 46. All methods use identical structure-matching, symmetry, and consensus criteria. Template retrieval is the strongest single method, reaching 68.3% top-1 success; symmetry-aware EquiCSP (66.4%) and Uni-3DAR (62.9%) form the next tier. However, comparison with TCSP 2.0 shows that most structures correctly predicted by generative models are also correctly predicted by template substitution. Thus, the set of structures uniquely reachable by generation is small, limiting its practical advantage for discovering structures outside existing prototype libraries. To test the source of this performance, we removed entire stoichiometric prototype families from the training set and retrained the strongest generative model. Accuracy declined by 50-78% across four families, establishing that performance is substantially prototype-dependent. A small minority of structures survived removal of their prototype family, demonstrating a real but limited retrieval-independent predictive capacity. Present generative CSP models therefore function largely as implicit, softer-edged prototype libraries rather than genuinely de novo predictors. Enlarging this residual capacity, rather than aggregate match rate alone, is the central open problem.

cond-mat.mtrl-sci↗
arXiv2026

GOTT: Object-centric Dexterous Manipulation with a Reusable Cross-Embodiment Primitive

Foundation models and large-scale human data provide rich sources of manipulation intent, but translating this intent into multi-fingered robot behavior remains difficult. Dexterous hands still lack a reusable low-level primitive that reliably establishes contact across tasks and embodiments. We propose GOTT, a reach-acquire-move framework built around a single cross-embodiment contact-acquisition primitive. Given a robot-agnostic object trajectory and a reach specification, GOTT first brings the hand near a task-relevant contact region. The shared closed-loop primitive then establishes stable contact from this approximate initialization, and a pose-conditioned controller tracks the desired object motion. Reach specifications may come from future-aware planning, external models, or human demonstrations, while the primitive and tracking backend remain unchanged. Simulation and real-world experiments show that GOTT is able to establish robust contact across diverse objects, arm-hand platforms, and seen and unseen hand morphologies. It also consistently improves end-to-end task success over open-loop grasp execution.

cs.RO↗
arXiv2026

SWE-Game: Can Coding Agents Build the Games We Want?

We introduce SWE-Game, a benchmark of 247 tasks grounded in 41 executable reference Godot games spanning 13 gameplay categories in 2D and 3D. Five task types cover development from a brief, implementation from a game design document, skeleton completion, repair of 83 injected-fault cases, and Godot-to-Unity porting. Reference materials specify the intended gameplay, while a shared instrumentation interface lets evaluator-owned drivers and probes execute actions and observe independently implemented games. Evaluation combines engine-state checks, certified reference-input replay, and agent-authored feature demonstrations to assess mechanic correctness, demonstrated playability, and behavioral restoration and preservation after repairs. Game-specific vision-language rubrics separately assess presentation. Across six models, Opus5 achieves the highest overall score in all five task types. Best overall scores remain below 60 out of 100 across the three construction tasks, with Brief-to-Game reaching 50.38. Analysis of reviewed submissions identifies requirement omissions and gameplay logic errors as predominant implementation problems. On human-labeled behaviors from 100 agent-built games, executable checks achieve 92.59% balanced accuracy, compared with 78.41% for a video-based VLM judge. Rubric-based visual scores reach a Spearman correlation of 0.829 with human ratings of 200 gameplay clips. Together, these results characterize current agent capabilities across game-development activities and support combining runtime evidence with visual assessment.

cs.AI↗
arXiv2026

uFlowCSP: Crystal Structure Prediction using Mean flow generative models

Crystal structure prediction (CSP) is fundamental to computational materials discovery. Generative models including CDVAE, DiffCSP, FlowMM, and CrystalFlow learn stable-crystal distributions directly, but diffusion and flow-matching inference requires tens to thousands of sequential network evaluations per candidate. We introduce uFlowCSP, a MeanFlow-based CSP model that learns the average, rather than instantaneous, probability-flow velocity. It generates a complete structure in one to five evaluations, delivering 5x-58x faster inference with equal or better performance. A chemistry- and symmetry-aware Transformer uses canonical atom ordering, global composition, and per-token chemistry embeddings. A coarse crystal-system token is used only during training; it provides additive gains, particularly improving space-group agreement despite being absent at inference, which remains formula-only. On MP-20 with 20 candidates per target, one step matches CrystalFlow (78.38% vs. 78.34%) with 100x fewer evaluations and about 10x lower wall-clock time. Five steps reach 83.64%, exceeding CrystalFlow (78.34% at 2,000 evaluations) and DiffCSP (77.93% at about 20,000), while using 20x fewer evaluations. uFlowCSP generates 10,000 structures in 0.39-1.31 minutes, versus 6.5 for CrystalFlow and 76.1 for DiffCSP. Under CSPBench's energy-ranked top-five structure-and-space-group criterion, five-step uFlowCSP reaches 72%/72%/65% structure, space-group, and consensus match rates. CrystalFlow reaches 78%/73%/68% at 100 steps but falls to 49%/32%/31% at five. Thus, uFlowCSP improves accuracy per network evaluation, not merely peak accuracy.

cond-mat.mtrl-sci↗
arXiv2026

The Illusion of Visual Tool-Use: A Causal Audit of Thinking with Images

The "thinking-with-images" paradigm equips multimodal LLMs with active visual operations such as crop-and-zoom. However, models using these operations often achieve only marginal or negative gains over direct inference at substantially higher token cost. They may also repeatedly crop irrelevant regions and fail on questions that direct inference answers correctly. We ask whether the returned visual evidence causally affects the answer. To answer this question, we formulate visual tool-use as a causal graph that separates observation-mediated paths from action-induced shortcuts. We then audit it through interventions at the three levels: policy (comparing tool-use with direct inference), trajectory (corrupting all observations during rollout), and step (counterfactually replacing one individual observation under a fixed prefix). Our step-level estimand, Visual Evidence Gain, isolates the contribution of each returned observation. Across six representative models and five fine-grained perception benchmarks, we uncover policy miscalibration with two failure modes. In Calling Without Looking, returned observations have no causal effect on the answer. In Looking Without Planning, observations are informative but the call schedule is incoherent. A trajectory-level diagnostic decomposes the policy-level accuracy gain into per-group contributions and shows that the gain is concentrated in a Calibrated minority. We term this discrepancy the illusion of visual tool-use: despite aggregate accuracy gains, visual tool-use is not causally effective across a broad range of rollouts. The code is available at https://github.com/OpenCausaLab/CauAudit.

cs.AI↗
arXiv2026

Attend to Evidence: Evidence-Anchored Spatial Attention Supervision for Multimodal RLVR

Reinforcement learning with verifiable rewards (RLVR) improves vision-language models (VLMs) by optimizing outcome rewards derived from final answers. However, such outcome-only rewards do not tell the model which image regions justify an answer. For questions that require visual grounding, these rewards cannot distinguish responses supported by relevant visual evidence from those produced by language-prior shortcuts or lucky guesses. We introduce EASE (Evidence-Anchored Spatial Attention), which augments multimodal RLVR with visual-evidence process supervision. EASE converts annotated evidence regions into a smoothed visual-token target and uses it to guide response-to-image attention during RL training, but only on high-reward trajectories. The annotations are used solely as privileged training labels, while inference requires only the original image and question. Across Qwen2.5-VL-7B, Qwen3-VL-4B, and Qwen3-VL-8B, EASE raises average scores over DAPO by 2.5 to 3.1 points on perception, hallucination, visual math, and multimodal reasoning benchmarks. Diagnostics and ablations show that EASE better aligns visual attention with annotated evidence regions.

cs.CV↗
arXiv2026

MedReaMM: Evaluating Large Multimodal Models on Expert-Level Clinical Diagnostic Synthesis

The application of Large Language Models (LLMs) to diagnostic decision-making has garnered growing interest. However, existing benchmarks largely focus on textual reasoning or isolated visual question-answering (VQA) tasks, lacking holistic integration of clinical narratives and medical imaging, and thus failing to assess the multimodal diagnostic synthesis capability central to expert clinical judgment. To bridge this gap, we introduce MedReaMM, a benchmark specifically designed to evaluate models' ability to synthesize heterogeneous clinical evidence consisting of detailed patient histories alongside multiple medical images into accurate differential diagnoses under a complete-information paradigm. Constructed from case reports sourced from top-tier medical journals and curated clinical case databases, MedReaMM comprises 625 expert-validated cases with an average of 2.79 medical images per case and a total of 1,042 standardized diagnoses annotated with ICD-11 codes. These cases predominantly represent rare, atypical, or multi-system presentations that demand expert-level evidence integration beyond routine pattern recognition. We evaluate 23 Large Multimodal Models (LMMs) and find that most achieve diagnostic accuracy scores below 50%, underscoring a substantial gap in multimodal diagnostic synthesis capability. Further analysis reveals that medical knowledge proficiency, medical image understanding, and evidence integration are all highly correlated with diagnostic performance.

cs.CV↗
arXiv2026

Enhancing Localized Reasoning for Long Video Understanding via Efficient Segment-to-Video Supervision

Though Multimodal Large Language Models (MLLMs) have shown impressive potential in video understanding, long video understanding (LVU) remains challenging since distracting noise in complex and lengthy contexts can obscure localized details, misleading MLLMs to produce incorrect answers. Recent works mitigate these issues by incentivizing deep reasoning to include relevant evidence. However, these methods have two main problems: First, the reinforcement fine-tuning framework (RFT) they leveraged incurs substantial training overheads, including high annotation costs and complicated reward designs. Second, the self-reflective and iterative-perception mechanism in some methods causes lengthy outputs and high inference latency. To alleviate these problems, we propose a novel Segment-to-Video Supervision} method (S2V) to efficiently enhance fine-grained reasoning in LVU. Specifically, we generate question answer pairs (VQA) based on localized segments, and then transfer these segment-based VQA back to the whole video for training. Due to focusing on short segments, segment-based VQA can naturally notice details which tend to be overlooked from a whole-video perspective. Training on such data can enforce MLLMs to correctly associate fine-grained details with QA while avoiding distracting noise in the whole video. The S2V training involves just reinforcement learning (RL) with a simple accuracy reward based on only 10K VQA samples and the resulting S2V model predicts answer using a single forward pass with limited output tokens. Experimental results demonstrate that S2V can consistently improve LVU performance across multiple LVU benchmarks, outperforming both general MLLMs and reasoning-based methods not only in LVU accuracy but also in training and inference efficiency.

cs.CV↗
arXiv2026

CLBench-V: Evaluating Multimodal Context Learning from Grounding to Knowledge Acquisition

Real-world tasks often require models to learn from task-specific context rather than relying only on pre-trained knowledge. While recent work has highlighted this capability as context learning, existing evaluations mainly focus on textual contexts. In many practical settings, however, the context to be learned from is multimodal: scientific findings are conveyed through figures and tables, financial indicators are scattered across converted reports, and spatial decisions depend on maps, scenes, or web pages. We introduce CLBench-V, a benchmark for multimodal context learning that addresses the difficulty of localizing where context use breaks down by organizing tasks around three dimensions: context grounding, new information application, and new knowledge learning. CLBench-V combines converted public benchmarks with newly constructed datasets spanning domains such as science, finance, long-document understanding, spatial reasoning, and web-based visual question answering. To reduce the cost of constructing domain-specific context-learning tasks, we further use automated construction and filtering procedures for our newly built datasets. Across 3,443 instances and six recent multimodal models, the best overall score is only 0.2847, indicating that multimodal context learning remains far from saturated. Moreover, InternVL3.5-30B-A3B performs best on context grounding and new knowledge learning, while Qwen3.5-Plus performs best on new information application. We further analyze judge reliability, context length, image count, and representative failure cases. Code is available at https://github.com/IamLihua/CLBench-V.

cs.CV↗
arXiv2026

Parameter Scan of Multi-Fluid Equilibria in Rotating p-11B Plasmas: Effects on Fusion Power and Bremsstrahlung Losses

We present VEQ-MF, a fast spectral parameter-scan framework for two-dimensional axisymmetric multi-fluid equilibria with prescribed species-dependent toroidal rotation. The solver couples generalized Boltzmann density responses, quasineutral electrostatic polarization, and a generalized Grad--Shafranov equation, extending reduced-parameter Grad-Shafranov and VEQ formulations to multi-species rotating equilibria. Rotating $p\text{-}^{11}\text{B}$ spherical-tokamak configurations are used as a demanding test case. Independent scans of the proton and boron rotation frequencies are performed in EHL-2 and EHL-3B geometries. The computed fields are then post-processed to obtain fusion power from a drift-Maxwellian reaction-rate coefficient and bremsstrahlung power from an analytical radiation model. Three in-range EHL-3B finite-difference benchmarks give global stored-energy, bremsstrahlung-power, and fusion-power differences of $1.7$--$3.4\%$, while a representative convergence check shows sub-percent sensitivity to increasing the spectral-parameter number and negligible sensitivity to Gaussian-grid refinement. The core equilibrium solve remains fast for repeated scans, with representative nonzero EHL-3B cases requiring $0.032$--$0.050$~s per point in MATLAB, excluding post-processing, interpolation, plotting, and file export. The scans identify two competing multi-fluid effects. Under iso-rotation, outward boron accumulation increases the volume-integrated $n_e^2$, so the fusion-to-bremsstrahlung power ratio $\mathcal{R}_{\mathrm{fb}}$ decreases with increasing rotation. Species-dependent toroidal rotation weakens centrifugal polarization and lowers bremsstrahlung power, while the relative toroidal flow in the larger EHL-3B geometry raises the drift-Maxwellian reaction-rate coefficient and thereby modifies fusion power.

physics.plasm-ph↗
arXiv2026

First reduced model for integrated computations of helicon wave heating and current drive in magnetic fusion plasmas

Fast predictive modelling of radio-frequency heating and current drive is important for integrated tokamak scenario design, yet kinetic calculations of helicon-wave absorption remain too computationally expensive for large-scale parameter scans. We present a reduced model for helicon-wave heating and current drive that retains the dominant parallel electron Landau-damping channel. The wave response is evaluated on the cold-plasma dispersion root, and a single-Landau-pole correction is introduced to obtain compact expressions for the local damping rate and current-drive efficiency. The model is benchmarked against the Chiu-Chan heating model using approximately 1.6 million samples covering representative conditions of EAST, HL-3, DIII-D and KSTAR. The reduction error is found to be governed primarily by the electron Landau parameter and electron beta. Within an identified sub-lower-hybrid-frequency validity window, results from different devices collapse onto a common error curve, which enables an empirical correction that is further tested using ITER-like and BEST-like extrapolation cases. Near and above the lower-hybrid frequency, the agreement deteriorates rapidly owing to changes in the cold-dispersion root structure and the breakdown of the single-branch WKB description. When coupled to a reduced current-drive source, the corrected heating model gives a median deviation of 10.8 percent from the Landau-channel Ehst-Karney reference and reproduces published CFETR current-density profiles. The resulting model provides a computationally efficient reduced closure for helicon-wave heating and current-drive calculations, together with physically interpretable limits on its range of validity.

physics.plasm-ph↗
arXiv2026

InvEvolve: Evolving White-Box Inventory Policies via Large Language Models with Performance Guarantees

We study how large language models can be used to generate inventory policies in online settings with non-stationary demand. Our work is motivated by recent advances in LLM-based evolutionary search, such as AlphaEvolve, which demonstrates strong performance on static and highly structured problems such as mathematical discovery, but is not directly suited to dynamic inventory settings with online updates. We propose InvEvolve, an end-to-end inventory policy evolution and inference framework grounded in confidence-interval-based certification. Built on a large language model trained via reinforcement learning, InvEvolve can process demand data together with additional numerical and textual features, and generates white-box inventory policies with statistical safety guarantees for future deployment. We further introduce a unified framework with theoretical guarantees that connects training, inference, and deployment. This allows us to derive a lower bound on the probability that InvEvolve evolves a statistically safe and improved policy, and to characterize the multi-period performance gap relative to the oracle-safe benchmark. Tested on both synthetic data and real-world retail data, InvEvolve outperforms classical inventory policies and deep-learning-based methods. In canonical inventory settings, it generates new policies that outperform existing benchmarks.

cs.LG↗
arXiv2026

CapSenseBand: Sustaining Cross-Disciplinary Creativity When Stitches Must Meet Signals

Wearable sensing systems increasingly depend on textiles that are both materially wearable and electronically functional. Their design requires collaboration between textile designers, who reason through stitches, yarn behavior, and machine constraints, and interaction designers, who reason through electrodes, signal paths, and insulation. However, these forms of expertise do not easily translate across disciplinary boundaries. This poster presents CapSenseBand, a knitted capacitive-sensing wristband developed through a research-through-design process organized around Analysis, Synthesis, and Detailing. We document an artifact chain spanning material swatches, a rapid wearable prototype, Paper Models as shared negotiation surfaces, a double-layer knitted structure, and an insulated Swept Frequency Capacitive Sensing breakout board. We show how Paper Models functioned as boundary objects, helping collaborators externalize intent, negotiate spatial and technical constraints, and preserve disciplinary expertise while converging on a shared design. We contribute a reusable swatch-to-sleeve pattern for material-centered HCI: keep discipline-specific probes open early, then converge through artifacts that make material, spatial, and electronic decisions legible before fabrication locks them in.

cs.HC↗
arXiv2026

Bridging What the Model Thinks and How It Speaks: Expressive Speech Generation via Self-Aware Intent-Realization Alignment

Speech Language Models (SLMs) exhibit strong semantic understanding, yet often fail to translate this capacity into expressive acoustic realization, producing speech with flattened prosody and misaligned emotion. We identify this mismatch as the semantic understanding-acoustic realization gap. Existing approaches typically rely on externally specified proxies, such as emotion labels or style prompts, which require annotations and struggle to capture dynamically evolving expressive intent throughout dialogue. To overcome these limitations, we propose SASLM (Self-Aware Speech Language Model), a proxy-free framework that bridges what the model thinks and how it speaks through self-aware intent-realization alignment: (1) Intent-Aware Bridging self-distills expressive intent from the model's own evolving semantic generation states via a Variational Information Bottleneck (VIB), thereby guiding expressive speech realization without external expressive supervision; while (2) Realization-Aware Alignment reflectively aligns generated acoustics with intended expression through self-reward optimization, progressively improving intent-realization consistency during speech generation. Despite using only 3B parameters and 800 hours of expressive speech data, SASLM achieves state-of-the-art performance on EchoMind among open-source systems, surpassing models over 10 times larger and approaching commercial systems.

cs.CL↗