arXiv subjects

Yilin Wu

Publications and source records attributed to Yilin Wu.

At least 19 recordsLinked to original sources

arXiv2026

Flow-polarity decoupling and universal mobility enhancement in dense bacterial active fluids with mesoscale order

Active fluids consisting of living cells or synthetic microswimmers display rich emergent behavior and nonequilibrium mechanical properties, which not only shed light on various biological processes but also inform the engineering of autonomous fluidics and self-driven materials. The individual behavior of microswimmers and their interaction with self-generated mesoscale solvent flows underlie the emergent properties of active fluids. Here we studied the microscopic dynamics in dense 3D bacterial active fluids by simultaneous imaging of cell body, flagella, and flow field. A surprising finding is that the polarity of cells was randomly distributed in mesoscale flow regimes, and yet the system displays mesoscale order in the self-generated solvent flows. Despite the apparent flow-polarity decoupling, the motion of cells relative to local solvent flows predominantly navigated upstream, with the self-advection speed universally enhanced by a flow-controlled constant. Numerical modeling with full hydrodynamic interactions reveals that the observed flow-polarity decoupling arises from the breakdown of the commonly held force-dipole assumption for anisotropic microswimmers: in the presence of flow gradient and near-field hydrodynamic interactions, the direction of total active forcing exerted by a swimming bacterium to the surrounding fluid no longer aligns with its polarity. The simulations suggest that near-field interactions serve as a new type of emergent, configuration-dependent active forcing, which profoundly impact self-organization and transport in dense bacterial suspensions. Taken together, our work establishes fundamental knowledge for faithfully understanding the collective behavior of dense polar active fluids.

physics.bio-ph↗

arXiv2026

WEAVER, Better, Faster, Longer: An Effective World Model for Robotic Manipulation

The potential impacts of world models (WMs, i.e., learned simulators) on robotics are far-reaching -- policy evaluation, policy improvement, and test-time planning -- all with limited real-world interaction. To unlock these downstream capabilities, a WM needs to jointly satisfy three desiderata: $\textit{(i)}$ fidelity (i.e., producing simulated trajectories that correlate with reality), $\textit{(ii)}$ consistency (i.e., producing simulated trajectories that are coherent over long horizons), and $\textit{(iii)}$ efficiency (i.e., producing simulated trajectories quickly). We propose WEAVER (World Estimation Across Views for Embodied Reasoning): a WM architecture that simultaneously achieves all three desiderata, providing state-of-the-art results on robotic manipulation tasks. WEAVER is a multi-view WM trained to predict future latents and reward values via a flow-matching loss. We distill the key design decisions across model architecture, memory, and prediction objectives required to unlock the kinds of long-horizon dynamic manipulation tasks that have confounded prior world modeling approaches. We apply WEAVER in robotic hardware, demonstrating its effectiveness at policy evaluation ($ρ$=0.870 correlation with real-world success rate), policy improvement (real-world success rate improvement of $38\%$ on top of the $π_{0.5}$ robot foundation model), and test-time planning (real-world success rate improvement of $14\%$ with a $5-10\times$ speedup over prior WMs). WEAVER also demonstrates better performance than prior WMs when evaluated on out-of-distribution scenarios. Code, models, and videos at: https://arnavkj1995.github.io/WEAVER/ .

cs.RO↗

arXiv2026

Inference-time Policy Steering via Vision and Touch

Inference-time steering adapts pre-trained generative robot policies during deployment by verifying candidate actions before execution. While prior methods typically perform this verification only with visual observations, vision alone is often insufficient for contact-rich manipulation, where success depends on both global task progress and subtle local interactions such as contact force. We introduce ViTaL, a visuo-tactile inference-time steering framework that formulates multimodal guidance as a bi-level optimization problem. At the high level, visual sampling-and-verification performs long-horizon mode selection, deciding what behavior the robot should execute. At the low level, tactile-guided diffusion editing refines the selected action sequence over a shorter horizon to satisfy local contact requirements. To support outcome-based steering, ViTaL learns a visuo-tactile latent world model and employs semantically aligned visual and tactile verifiers, including a novel text-conditioned tactile reward that scores predicted tactile futures directly in latent space. Across three real-world contact-rich manipulation tasks, ViTaL improves overall success by 51% over the base policy, outperforms unimodal steering by at least 33%, and exceeds naive multimodal fusion by at least 20%. Website: https://yilin-wu98.github.io/vital_website.

cs.RO↗

arXiv2026

Position: Good Embodied Reward Models Need Bad Behavior Data

This position paper argues that to obtain reliable embodied reward models, the community must invest in ``bad'' robot data: failed, suboptimal, error-prone, and even hazardous behaviors. While reward models are central to any foundation model's lifecycle, today's embodied reward models are trained primarily on successful behaviors. We analyze three state-of-the-art embodied reward models and find that they systematically over-reward behaviors that real human evaluators would penalize, including unsafe interactions, poor execution, and shortcut strategies that only superficially satisfy tasks. We attribute these failures to a key data gap: the scarcity of negative embodied data which is costly to collect and often filtered out or withheld in existing robotics datasets. Furthermore, we show that even modest exposure to real bad behavior data can improve alignment with human preferences and reduce costly false positives. We therefore call on the embodied AI community to curate and release their bad robot data, build synthetic bad data generation engines, develop more decentralized physical evaluation systems, and design benchmarks for fine-grained embodied reward model evaluations.

cs.RO↗

arXiv2026

When to Act, Ask, or Learn: Uncertainty-Aware Policy Steering

Policy steering is an emerging way to adapt robot behaviors at deployment-time: a learned verifier analyzes low-level action samples proposed by a pre-trained policy (e.g., diffusion policy) and selects only those aligned with the task. While Vision-Language Models (VLMs) are promising general-purpose verifiers due to their reasoning capabilities, existing frameworks often assume these models are well-calibrated. In practice, the overconfident judgment from VLM can degrade the steering performance under both high-level semantic uncertainty in task specifications and low-level action uncertainty or incapability of the pre-trained policy. We propose uncertainty-aware policy steering (UPS), a framework that jointly reasons about semantic task uncertainty and low-level action feasibility, and selects an uncertainty resolution strategy: execute a high-confidence action, clarify task ambiguity via natural language queries, or ask for action interventions to correct the low-level policy when it is deemed incapable at the task. We leverage conformal prediction to calibrate the composition of the VLM and the pre-trained base policy, providing statistical assurances that the verifier selects the correct strategy. After collecting interventions during deployment, we employ residual learning to improve the capability of the pre-trained policy, enabling the system to learn continually but with minimal expensive human feedback. We demonstrate our framework through experiments in simulation and on hardware, showing that UPS can disentangle confident, ambiguous, and incapable scenarios and minimizes expensive user interventions compared to uncalibrated baselines and prior human- or robot-gated continual learning approaches. Videos can be found at https://jessie-yuan.github.io/ups/

cs.RO↗

arXiv2026

Group actions on relative cluster categories and Higgs categories

Let $G$ be a finite group acting on an ice quiver with potential $(Q, F, W)$. We construct the corresponding $G$-equivariant relative cluster category and $G$-equivariant Higgs category, extending the work of Demonet. Using the orbit mutations on the set of $G$-stable cluster-tilting objects of the Higgs category and an appropriate cluster character, we can link these data to an explicit skew-symmetrizable cluster algebra with coefficients. As a specific example, this provides an additive categorification for cluster algebras with principal coefficients in the non-simply laced case.

math.RT↗

arXiv2026

Silting reduction, relative AGK's construction and Higgs construction

We introduce the notion of a Calabi--Yau quadruple as a generalization of Iyama--Yang's Calabi--Yau triple. For each $(d+1)$-Calabi--Yau quadruple, we show that the associated Higgs category is a $d$-Calabi--Yau Frobenius extriangulated category, which moreover admits a canonical $d$-cluster-tilting subcategory. Concrete examples arise from the construction of relative cluster categories and Higgs categories in the setting of ice quivers with potentials, as well as from the singularity category of an isolated singularity. As an application, we prove that both the relative Amiot--Guo--Keller's construction and the Higgs construction of a $(d+1)$-Calabi--Yau quadruple take silting reduction to Calabi--Yau reduction.

math.RT↗

arXiv2026

Bell Plesset Effects on Rayleigh Taylor Instability of Three Dimensional Spherical Geometry

We develop a weakly nonlinear, multi-mode theory for the Rayleigh-Taylor instability (RTI) on a time-varying spherical interface, fully incorporating mode couplings and the Bell-Plesset (BP) effects arising from interface convergence. Our model extends prior analyses, which have been largely restricted to static backgrounds, 2D cylindrical geometries, or single-mode initial condition. We present a framework capable of evolving arbitrary, fully three-dimensional initial perturbations on a dynamic background. At the first order, mode amplitudes respond to the time-varying interface acceleration with an exponential-like growth, in qualitative agreement with classic static results. At second order, nonlinear mode coupling reveals a powerful selection rule: energy is preferentially channeled into axisymmetric (m=0) modes. We find that the BP effects dramatically amplify the instability growth by a few orders of magnitude, with this amplification being even more significant for second order couplings. Despite this strong channeling, the second order amplitudes remain small relative to the first order, validating the perturbative approach. These findings offer new physical insights into time-dependent interface instabilities relevant to applications such as astrophysical shell collapse and inertial confinement fusion, highlighting the uniquely dominant role of axisymmetric modes in BP-driven convergent flows.

physics.plasm-ph↗

arXiv2026

Do What You Say: Steering Vision-Language-Action Models via Runtime Reasoning-Action Alignment Verification

Reasoning Vision Language Action (VLA) models improve robotic instruction-following by generating step-by-step textual plans before low-level actions, an approach inspired by Chain-of-Thought (CoT) reasoning in language models. Yet even with a correct textual plan, the generated actions can still miss the intended outcomes in the plan, especially in out-of-distribution (OOD) scenarios. We formalize this phenomenon as a lack of embodied CoT faithfulness, and introduce a training-free, runtime policy steering method for reasoning-action alignment. Given a reasoning VLA's intermediate textual plan, our framework samples multiple candidate action sequences from the same model, predicts their outcomes via simulation, and uses a pre-trained Vision-Language Model (VLM) to select the sequence whose outcome best aligns with the VLA's own textual plan. Only executing action sequences that align with the textual reasoning turns our base VLA's natural action diversity from a source of error into a strength, boosting robustness to semantic and visual OOD perturbations and enabling novel behavior composition without costly re-training. We also contribute a reasoning-annotated extension of LIBERO-100, environment variations tailored for OOD evaluation, and demonstrate up to 15% performance gain over prior work on behavior composition tasks and scales with compute and data diversity. Project Website at: https://yilin-wu98.github.io/steering-reasoning-vla/

cs.RO↗

arXiv2026

Empowering Small Language Models with Factual Hallucination-Aware Reasoning for Financial Classification

Small language models (SLMs) are increasingly used for financial classification due to their fast inference and local deployability. However, compared with large language models, SLMs are more prone to factual hallucinations in reasoning and exhibit weaker classification performance. This raises a natural question: Can mitigating factual hallucinations improve SLMs' financial classification? To address this, we propose a three-step pipeline named AAAI (Association Identification, Automated Detection, and Adaptive Inference). Experiments on three representative SLMs reveal that: (1) factual hallucinations are positively correlated with misclassifications; (2) encoder-based verifiers effectively detect factual hallucinations; and (3) incorporating feedback on factual errors enables SLMs' adaptive inference that enhances classification performance. We hope this pipeline contributes to trustworthy and effective applications of SLMs in finance.

cs.AI↗

arXiv2025

PanoNav: Mapless Zero-Shot Object Navigation with Panoramic Scene Parsing and Dynamic Memory

Zero-shot object navigation (ZSON) in unseen environments remains a challenging problem for household robots, requiring strong perceptual understanding and decision-making capabilities. While recent methods leverage metric maps and Large Language Models (LLMs), they often depend on depth sensors or prebuilt maps, limiting the spatial reasoning ability of Multimodal Large Language Models (MLLMs). Mapless ZSON approaches have emerged to address this, but they typically make short-sighted decisions, leading to local deadlocks due to a lack of historical context. We propose PanoNav, a fully RGB-only, mapless ZSON framework that integrates a Panoramic Scene Parsing module to unlock the spatial parsing potential of MLLMs from panoramic RGB inputs, and a Memory-guided Decision-Making mechanism enhanced by a Dynamic Bounded Memory Queue to incorporate exploration history and avoid local deadlocks. Experiments on the public navigation benchmark show that PanoNav significantly outperforms representative baselines in both SR and SPL metrics.

cs.CV↗

arXiv2025

Nonequilibrium dynamics of membraneless active droplets

Membraneless droplets or liquid condensates formed via liquid-liquid phase separation (LLPS) play a pivotal role in cell biology and hold potential for biomedical engineering. While membraneless droplets are often studied in the context of interactions between passive components, it is increasingly recognized that active matter inclusions, such as molecular motors and catalytic enzymes in cells, play important roles in the formation, transport and interaction of membraneless droplets. Here we developed a bacteria-polymer active phase separation system to study the nonequilibrium effect of active matter inclusions on the LLPS dynamics. We found that the presence of bacterial active matter accelerated the initial condensation of phase-separated liquid droplets but subsequently arrested the droplet coarsening process, resulting in a stable suspension of membraneless active droplets packed with motile bacterial cells. The arrested phase separation of the bacterial active droplet system presumably arises from anti-phase entrainment of interface fluctuations between neighboring droplets, which reduces the frequency of inter-droplet contact and suppresses droplet coarsening. In addition, the active stresses generated by cells within the droplets give rise to an array of nonequilibrium phenomena, such as dominant long-wavelength fluctuations and enhanced droplet transport with short-term persistent motion due to spontaneous symmetry breaking. Our study reveals a unique mechanism for arrested phase separation and long-term stability in membraneless droplet systems. The bacteria-polymer active phase separation system opens a new avenue for studying the dynamics of membraneless active droplets relevant to non-equilibrium LLPS in cells and in biomedical engineering applications.

cond-mat.soft↗

arXiv2025

Artificial ferroelectric-like hysteresis in antiferroelectrics with non-uniform disorder

Antiferroelectrics exhibit unique double-hysteresis polarization loops, which have garnered significant attention due to their potential applications such as energy storage, electromechanical transduction, as well as synapse devices. However, numerous antiferroelectric materials have been reported to display signs of hysteresis loops resembling those of ferroelectric materials, and a comprehensive understanding remains elusive. In this work, we provide a phenomenological model that reproduces such widely observed artificial ferroelectric hysteresis with a superposition of numerous disordered antiferroelectric loops that have varying antiferroelectric-to-ferroelectric transition fields, particularly when these field ranges intersect. Experimentally, we realized such artificial ferroelectric-like hysteresis loops in the prototypical antiferroelectric PbZrO$_3$ and PbHfO$_3$ thin films, by introducing non-uniform local disorder (e.g., defects) via fine-tuning of the film growth conditions. These ferroelectric-like states are capable of persisting for several hours prior to transitioning back into the thermodynamically stable antiferroelectric ground state. Those results provide insights into the fundamental impact of disorder on the AFE properties and new possibilities of disorder-tailored functions.

cond-mat.mtrl-sci↗

arXiv2025

Etching-free dual-lift-off for direct patterning of epitaxial oxide thin films

Although monocrystalline oxide films offer broad functional capabilities, their practical use is hampered by challenges in patterning. Traditional patterning relies on etching, which can be costly and prone to issues like film or substrate damage, under-etching, over-etching, and lateral etching. In this study, we introduce a dual-lift-off method for direct patterning of oxide films, circumventing the etching process and associated issues. Our method involves an initial lift-off of amorphous Sr$_3$Al$_2$O$_6$ or Sr$_4$Al$_2$O$_7$ ($a$SAO) through stripping the photoresist, followed by a subsequent lift-off of the functional oxide thin films by dissolving the $a$SAO layer. $a$SAO functions as a ``high-temperature photoresist", making it compatible with the high-temperature growth of monocrystalline oxides. Using this method, patterned ferromagnetic La$_{0.67}$Sr$_{0.33}$MnO$_{3}$ and ferroelectric BiFeO$_3$ were fabricated, accurately mirroring the shape of the photoresist. Our study presents a straightforward, flexible, precise, environmentally friendly, and cost-effective method for patterning high-quality oxide thin films.

cond-mat.mtrl-sci↗

arXiv2025

LargeMvC-Net: Anchor-based Deep Unfolding Network for Large-scale Multi-view Clustering

Deep anchor-based multi-view clustering methods enhance the scalability of neural networks by utilizing representative anchors to reduce the computational complexity of large-scale clustering. Despite their scalability advantages, existing approaches often incorporate anchor structures in a heuristic or task-agnostic manner, either through post-hoc graph construction or as auxiliary components for message passing. Such designs overlook the core structural demands of anchor-based clustering, neglecting key optimization principles. To bridge this gap, we revisit the underlying optimization problem of large-scale anchor-based multi-view clustering and unfold its iterative solution into a novel deep network architecture, termed LargeMvC-Net. The proposed model decomposes the anchor-based clustering process into three modules: RepresentModule, NoiseModule, and AnchorModule, corresponding to representation learning, noise suppression, and anchor indicator estimation. Each module is derived by unfolding a step of the original optimization procedure into a dedicated network component, providing structural clarity and optimization traceability. In addition, an unsupervised reconstruction loss aligns each view with the anchor-induced latent space, encouraging consistent clustering structures across views. Extensive experiments on several large-scale multi-view benchmarks show that LargeMvC-Net consistently outperforms state-of-the-art methods in terms of both effectiveness and scalability.

cs.CV↗

arXiv2025

From Foresight to Forethought: VLM-In-the-Loop Policy Steering via Latent Alignment

While generative robot policies have demonstrated significant potential in learning complex, multimodal behaviors from demonstrations, they still exhibit diverse failures at deployment-time. Policy steering offers an elegant solution to reducing the chance of failure by using an external verifier to select from low-level actions proposed by an imperfect generative policy. Here, one might hope to use a Vision Language Model (VLM) as a verifier, leveraging its open-world reasoning capabilities. However, off-the-shelf VLMs struggle to understand the consequences of low-level robot actions as they are represented fundamentally differently than the text and images the VLM was trained on. In response, we propose FOREWARN, a novel framework to unlock the potential of VLMs as open-vocabulary verifiers for runtime policy steering. Our key idea is to decouple the VLM's burden of predicting action outcomes (foresight) from evaluation (forethought). For foresight, we leverage a latent world model to imagine future latent states given diverse low-level action plans. For forethought, we align the VLM with these predicted latent states to reason about the consequences of actions in its native representation--natural language--and effectively filter proposed plans. We validate our framework across diverse robotic manipulation tasks, demonstrating its ability to bridge representational gaps and provide robust, generalizable policy steering. Videos can be found on the project website: https://yilin-wu98.github.io/forewarn/.

cs.RO↗

arXiv2025

DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset

The creation of large, diverse, high-quality robot manipulation datasets is an important stepping stone on the path toward more capable and robust robotic manipulation policies. However, creating such datasets is challenging: collecting robot manipulation data in diverse environments poses logistical and safety challenges and requires substantial investments in hardware and human labour. As a result, even the most general robot manipulation policies today are mostly trained on data collected in a small number of environments with limited scene and task diversity. In this work, we introduce DROID (Distributed Robot Interaction Dataset), a diverse robot manipulation dataset with 76k demonstration trajectories or 350 hours of interaction data, collected across 564 scenes and 84 tasks by 50 data collectors in North America, Asia, and Europe over the course of 12 months. We demonstrate that training with DROID leads to policies with higher performance and improved generalization ability. We open source the full dataset, policy learning code, and a detailed guide for reproducing our robot hardware setup.

cs.RO↗

arXiv2024

Maximizing Alignment with Minimal Feedback: Efficiently Learning Rewards for Visuomotor Robot Policy Alignment

Visuomotor robot policies, increasingly pre-trained on large-scale datasets, promise significant advancements across robotics domains. However, aligning these policies with end-user preferences remains a challenge, particularly when the preferences are hard to specify. While reinforcement learning from human feedback (RLHF) has become the predominant mechanism for alignment in non-embodied domains like large language models, it has not seen the same success in aligning visuomotor policies due to the prohibitive amount of human feedback required to learn visual reward functions. To address this limitation, we propose Representation-Aligned Preference-based Learning (RAPL), an observation-only method for learning visual rewards from significantly less human preference feedback. Unlike traditional RLHF, RAPL focuses human feedback on fine-tuning pre-trained vision encoders to align with the end-user's visual representation and then constructs a dense visual reward via feature matching in this aligned representation space. We first validate RAPL through simulation experiments in the X-Magical benchmark and Franka Panda robotic manipulation, demonstrating that it can learn rewards aligned with human preferences, more efficiently uses preference data, and generalizes across robot embodiments. Finally, our hardware experiments align pre-trained Diffusion Policies for three object manipulation tasks. We find that RAPL can fine-tune these policies with 5x less real human preference data, taking the first step towards minimizing human feedback while maximizing visuomotor robot policy alignment.

cs.RO↗