arXiv subjects

Yilin Wu

Publications and source records attributed to Yilin Wu.

At least 37 records · Page 2Linked to original sources

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↗

arXiv2024

FinRobot: AI Agent for Equity Research and Valuation with Large Language Models

As financial markets grow increasingly complex, there is a rising need for automated tools that can effectively assist human analysts in equity research, particularly within sell-side research. While Generative AI (GenAI) has attracted significant attention in this field, existing AI solutions often fall short due to their narrow focus on technical factors and limited capacity for discretionary judgment. These limitations hinder their ability to adapt to new data in real-time and accurately assess risks, which diminishes their practical value for investors. This paper presents FinRobot, the first AI agent framework specifically designed for equity research. FinRobot employs a multi-agent Chain of Thought (CoT) system, integrating both quantitative and qualitative analyses to emulate the comprehensive reasoning of a human analyst. The system is structured around three specialized agents: the Data-CoT Agent, which aggregates diverse data sources for robust financial integration; the Concept-CoT Agent, which mimics an analysts reasoning to generate actionable insights; and the Thesis-CoT Agent, which synthesizes these insights into a coherent investment thesis and report. FinRobot provides thorough company analysis supported by precise numerical data, industry-appropriate valuation metrics, and realistic risk assessments. Its dynamically updatable data pipeline ensures that research remains timely and relevant, adapting seamlessly to new financial information. Unlike existing automated research tools, such as CapitalCube and Wright Reports, FinRobot delivers insights comparable to those produced by major brokerage firms and fundamental research vendors. We open-source FinRobot at \url{https://github. com/AI4Finance-Foundation/FinRobot}.

q-fin.CP↗

arXiv2024

Learning Generalizable Tool-use Skills through Trajectory Generation

Autonomous systems that efficiently utilize tools can assist humans in completing many common tasks such as cooking and cleaning. However, current systems fall short of matching human-level of intelligence in terms of adapting to novel tools. Prior works based on affordance often make strong assumptions about the environments and cannot scale to more complex, contact-rich tasks. In this work, we tackle this challenge and explore how agents can learn to use previously unseen tools to manipulate deformable objects. We propose to learn a generative model of the tool-use trajectories as a sequence of tool point clouds, which generalizes to different tool shapes. Given any novel tool, we first generate a tool-use trajectory and then optimize the sequence of tool poses to align with the generated trajectory. We train a single model on four different challenging deformable object manipulation tasks, using demonstration data from only one tool per task. The model generalizes to various novel tools, significantly outperforming baselines. We further test our trained policy in the real world with unseen tools, where it achieves the performance comparable to human. Additional materials can be found on our project website: https://sites.google.com/view/toolgen.

cs.RO↗

arXiv2024

Giant enhancement of bacterial upstream swimming in macromolecular flows

Many bacteria live in natural and clinical environments with abundant macromolecular polymers. Macromolecular fluids commonly display viscoelasticity and non-Newtonian rheological behavior; it is unclear how these complex-fluid properties affect bacterial transport in flows. Here we combine high-resolution microscopy and numerical simulations to study bacterial response to shear flows of various macromolecular fluids. In stark contrast to the case in Newtonian shear flows, we found that flagellated bacteria in macromolecular flows display a giant capacity of upstream swimming (a behavior resembling fish swimming against current) near solid surfaces: The cells can counteract flow washing at shear rates up to ~65 $s^{-1}$, one order of magnitude higher than the limit for cells swimming in Newtonian flows. The significant enhancement of upstream swimming depends on two characteristic complex-fluid properties, namely viscoelasticity and shear-thinning viscosity; meanwhile, increasing the viscosity with a Newtonian polymer can prevent upstream motion. By visualizing flagellar bundles and modeling bacterial swimming in complex fluids, we explain the phenomenon as primarily arising from the augmentation of a "weathervane effect" in macromolecular flows due to the presence of a viscoelastic lift force and a shear-thinning induced azimuthal torque promoting the alignment of bacteria against the flow direction. Our findings shed light on bacterial transport and surface colonization in macromolecular environments, and may inform the design of artificial helical microswimmers for biomedical applications in physiological conditions.

cond-mat.soft↗

arXiv2024

HACMan++: Spatially-Grounded Motion Primitives for Manipulation

Although end-to-end robot learning has shown some success for robot manipulation, the learned policies are often not sufficiently robust to variations in object pose or geometry. To improve the policy generalization, we introduce spatially-grounded parameterized motion primitives in our method HACMan++. Specifically, we propose an action representation consisting of three components: what primitive type (such as grasp or push) to execute, where the primitive will be grounded (e.g. where the gripper will make contact with the world), and how the primitive motion is executed, such as parameters specifying the push direction or grasp orientation. These three components define a novel discrete-continuous action space for reinforcement learning. Our framework enables robot agents to learn to chain diverse motion primitives together and select appropriate primitive parameters to complete long-horizon manipulation tasks. By grounding the primitives on a spatial location in the environment, our method is able to effectively generalize across object shape and pose variations. Our approach significantly outperforms existing methods, particularly in complex scenarios demanding both high-level sequential reasoning and object generalization. With zero-shot sim-to-real transfer, our policy succeeds in challenging real-world manipulation tasks, with generalization to unseen objects. Videos can be found on the project website: https://sgmp-rss2024.github.io.

cs.RO↗

arXiv2024

Emergence of large-scale mechanical spiral waves in bacterial living matter

Propagating spiral waves have been discovered in various chemical, biological and physical systems. Spiral waves in multicellular organisms are often associated with essential living functions. Although certain eukaryotic microorganisms have long been known to generate spiral waves, evidence of spiral wave pattern has not been lacking in the bacterial world. Here we report the discovery of a unique form of propagating spiral waves in dense bacterial populations where cells engage in cyclic force-generating processes driven by type-IV pilus motility. Specifically, we discovered that synchronization of pilus activity in the bacterial living matter leads to large-scale spatiotemporal regulation of tension force in the form of propagating spiral waves. Theoretical modelling reveals that the spiral tension waves result from nonreciprocity in cell-cell interactions. Our findings reveal a novel mechanism of large-scale force regulation in bacterial world and may shed light on the emergent mechanics of biofilms and microbiomes. Pilus-driven bacterial living matter also provides a mechanical active medium for studying electrical or chemical spiral waves in living systems.

physics.bio-ph↗

arXiv2024

Self-enhanced mobility enables vortex pattern formation in living matter

Emergence of regular spatial patterns is a hallmark in living matter ranging from subcellular organelles to developing embryos and to ecosystems. Mechanisms for the formation of ordered spatial patterns in biology often require chemical signaling that coordinates cellular behavior and differentiation. Here we discovered a novel route to large-scale regular pattern formation in living matter mediated by purely physical interactions. We found that dense bacterial living matter spontaneously developed an ordered lattice of mesoscale, fast-spinning vortices each consisting of ~10^4-10^5 motile cells; these mesoscale vortices were arranged in space over centimeter scale with apparent hexagonal order, while individual cells in the vortices moved in coordinated directions with strong polar and vortical order. Single-cell tracking and numerical simulations suggest that the phenomenon is enabled by self-enhanced mobility of individual cells in the system. Our findings demonstrate a simple physical mechanism for self-organized pattern formation in living systems and more generally, in other active matter systems near the boundary of fluidlike and solidlike behaviors.

physics.bio-ph↗

arXiv2023

Relative cluster categories and Higgs categories with infinite-dimensional morphism spaces

Cluster algebras *with coefficients* are important since they appear in nature as coordinate algebras of varieties like Grassmannians, double Bruhat cells, unipotent cells, ... . The approach of Geiss-Leclerc-Schröer often yields Frobenius exact categories which allow to categorify such cluster algebras. In previous work, the third-named author has constructed Higgs categories and relative cluster categories in the relative Jacobi-finite setting (arXiv:2109.03707). Higgs categories generalize the Frobenius categories used by Geiss-Leclerc-Schröer. In this article, we construct the Higgs category and the relative cluster category in the relative Jacobi-infinite setting under suitable hypotheses. These cover for example the case of Jensen-King-Su's Grassmannian cluster category. As in the relative Jacobi-finite case, the Higgs category is no longer exact but still extriangulated in the sense of Nakaoka-Palu. We also construct a cluster character refining Plamondon's. In the appendix, Chris Fraser and the second-named author categorify quasi-cluster morphisms using Frobenius categories. A recent application of this result is due to Matthew Pressland, who uses it to prove a conjecture by Muller-Speyer.

math.RT↗

arXiv2023

Stabilize to Act: Learning to Coordinate for Bimanual Manipulation

Key to rich, dexterous manipulation in the real world is the ability to coordinate control across two hands. However, while the promise afforded by bimanual robotic systems is immense, constructing control policies for dual arm autonomous systems brings inherent difficulties. One such difficulty is the high-dimensionality of the bimanual action space, which adds complexity to both model-based and data-driven methods. We counteract this challenge by drawing inspiration from humans to propose a novel role assignment framework: a stabilizing arm holds an object in place to simplify the environment while an acting arm executes the task. We instantiate this framework with BimanUal Dexterity from Stabilization (BUDS), which uses a learned restabilizing classifier to alternate between updating a learned stabilization position to keep the environment unchanged, and accomplishing the task with an acting policy learned from demonstrations. We evaluate BUDS on four bimanual tasks of varying complexities on real-world robots, such as zipping jackets and cutting vegetables. Given only 20 demonstrations, BUDS achieves 76.9% task success across our task suite, and generalizes to out-of-distribution objects within a class with a 52.7% success rate. BUDS is 56.0% more successful than an unstructured baseline that instead learns a BC stabilizing policy due to the precision required of these complex tasks. Supplementary material and videos can be found at https://sites.google.com/view/stabilizetoact .

cs.RO↗

arXiv2023

Relative cluster categories and Higgs categories

Cluster categories were introduced in 2006 by Buan-Marsh-Reineke-Reiten-Todorov in order to categorify acyclic cluster algebras without coefficients. Their construction was generalized by Amiot (2009) and Plamondon (2011) to arbitrary cluster algebras associated with quivers. A higher dimensional generalization is due to Guo (2011). Cluster algebras with coefficients are important since they appear in nature as coordinate algebras of varieties like Grassmannians, double Bruhat cells, unipotent cells, etc. The work of Geiss-Leclerc-Schröer often yields Frobenius exact categories which allow us to categorify such cluster algebras. In this paper, we generalize the construction of (higher) cluster categories by Claire Amiot and by Lingyan Guo to the relative context. We prove the existence of an $ n $-cluster tilting object in a Frobenius extriangulated category, namely the Higgs category (generalizing the Frobenius categories of Geiss-Leclerc-Schröer), which is stably $ n $-Calabi--Yau and Hom-finite, arising from a left $ (n+1) $-Calabi--Yau morphism. Our results apply in particular to relative Ginzburg dg algebras coming from ice quivers with potential and higher Auslander algebras associated to $ n $-representation-finite algebras.

math.RT↗

arXiv2023

Categorification of ice quiver mutation

In 2009, Keller and Yang categorified quiver mutation by interpreting it in terms of equivalences between derived categories. Their approach was based on Ginzburg's Calabi-Yau algebras and on Derksen-Weyman-Zelevinsky's mutation of quivers with potential. Recently, Matthew Pressland has generalized mutation of quivers with potential to that of ice quivers with potential. In this paper, we show that his rule yields derived equivalences between the associated relative Ginzburg algebras, which are special cases of Yeung's deformed relative Calabi-Yau completions arising in the theory of relative Calabi-Yau structures due to Toën and Brav-Dyckerhoff. We illustrate our results on examples arising in the work of Baur-King-Marsh on dimer models and cluster categories of Grassmannians. We also give a categorification of mutation at frozen vertices as it appears in recent work of Fraser-Sherman-Bennett on positroid cluster structures.

math.RT↗

arXiv2023

In-Mouth Robotic Bite Transfer with Visual and Haptic Sensing

Assistance during eating is essential for those with severe mobility issues or eating risks. However, dependence on traditional human caregivers is linked to malnutrition, weight loss, and low self-esteem. For those who require eating assistance, a semi-autonomous robotic platform can provide independence and a healthier lifestyle. We demonstrate an essential capability of this platform: safe, comfortable, and effective transfer of a bite-sized food item from a utensil directly to the inside of a person's mouth. Our system uses a force-reactive controller to safely accommodate the user's motions throughout the transfer, allowing full reactivity until bite detection then reducing reactivity in the direction of exit. Additionally, we introduce a novel dexterous wrist-like end effector capable of small, unimposing movements to reduce user discomfort. We conduct a user study with 11 participants covering 8 diverse food categories to evaluate our system end-to-end, and we find that users strongly prefer our method to a wide range of baselines. Appendices and videos are available at our website: https://tinyurl.com/btICRA.

cs.RO↗

arXiv2022

Learning Bimanual Scooping Policies for Food Acquisition

A robotic feeding system must be able to acquire a variety of foods. Prior bite acquisition works consider single-arm spoon scooping or fork skewering, which do not generalize to foods with complex geometries and deformabilities. For example, when acquiring a group of peas, skewering could smoosh the peas while scooping without a barrier could result in chasing the peas on the plate. In order to acquire foods with such diverse properties, we propose stabilizing food items during scooping using a second arm, for example, by pushing peas against the spoon with a flat surface to prevent dispersion. The added stabilizing arm can lead to new challenges. Critically, this arm should stabilize the food scene without interfering with the acquisition motion, which is especially difficult for easily breakable high-risk food items like tofu. These high-risk foods can break between the pusher and spoon during scooping, which can lead to food waste falling out of the spoon. We propose a general bimanual scooping primitive and an adaptive stabilization strategy that enables successful acquisition of a diverse set of food geometries and physical properties. Our approach, CARBS: Coordinated Acquisition with Reactive Bimanual Scooping, learns to stabilize without impeding task progress by identifying high-risk foods and robustly scooping them using closed-loop visual feedback. We find that CARBS is able to generalize across food shape, size, and deformability and is additionally able to manipulate multiple food items simultaneously. CARBS achieves 87.0% success on scooping rigid foods, which is 25.8% more successful than a single-arm baseline, and reduces food breakage by 16.2% compared to an analytical baseline. Videos can be found at https://sites.google.com/view/bimanualscoop-corl22/home .

cs.RO↗

arXiv2022

Geometrical control of interface patterning underlies active matter invasion

Interaction between active materials and the boundaries of geometrical confinement is key to many emergent phenomena in active systems. For living active matter consisting of animal cells or motile bacteria, the confinement boundary is often a deformable interface, and it has been unclear how activity-induced interface dynamics might lead to morphogenesis and pattern formation. Here we studied the evolution of bacterial active matter confined by a deformable boundary. We discovered that an ordered morphological pattern emerged at the interface characterized by periodically-spaced interfacial protrusions; behind the interfacial protrusions, bacterial swimmers self-organized into multicellular clusters displaying +1/2 nematic defects. Subsequently, a hierarchical sequence of transitions from interfacial protrusions to creeping branches allowed the bacterial active drop to rapidly invade surrounding space with a striking self-similar branch pattern. We found that this interface patterning is controlled by the local curvature of the interface, a phenomenon we denote as collective curvature sensing. Using a continuum active model, we revealed that the collective curvature sensing arises from enhanced active stresses near high-curvature regions, with the active length-scale setting the characteristic distance between the interfacial protrusions. Our findings reveal a protrusion-to-branch transition as a novel mode of active matter invasion and suggest a new strategy to engineer pattern formation of active materials.

cond-mat.soft↗

arXiv2022

Autonomous waves and global motion modes in living active solids

Elastic active matter or active solid consists of self-propelled units embedded in an elastic matrix. Active solid resists deformation; the shape-preserving property and the intrinsic non-equilibrium nature make active solids a superior component for self-driven devices. Nonetheless, the mechanical properties and emergent behavior of active solids are poorly understood. Using a biofilm-based bacterial active solid, here we discovered self-sustained elastic waves with unique wave properties not seen in passive solids, such as power-law scaling of wave speed with activity. Under isotropic confinement, the active solid develops two topologically distinct global motion modes that can be selectively excited, with a surprising step-like frequency jump at mode transition. Our findings reveal novel spatiotemporal order in elastic active matter and may guide the development of solid-state adaptive or living materials.

cond-mat.soft↗

arXiv2021

Solving Compositional Reinforcement Learning Problems via Task Reduction

We propose a novel learning paradigm, Self-Imitation via Reduction (SIR), for solving compositional reinforcement learning problems. SIR is based on two core ideas: task reduction and self-imitation. Task reduction tackles a hard-to-solve task by actively reducing it to an easier task whose solution is known by the RL agent. Once the original hard task is successfully solved by task reduction, the agent naturally obtains a self-generated solution trajectory to imitate. By continuously collecting and imitating such demonstrations, the agent is able to progressively expand the solved subspace in the entire task space. Experiment results show that SIR can significantly accelerate and improve learning on a variety of challenging sparse-reward continuous-control problems with compositional structures. Code and videos are available at https://sites.google.com/view/sir-compositional.

cs.LG↗

arXiv2020

Viscoelastic control of spatiotemporal order in bacterial active matter

Active matter consists of units that generate mechanical work by consuming energy. Examples include living systems, such as assemblies of bacteria and biological tissues, biopolymers driven by molecular motors, and suspensions of synthetic self-propelled particles. A central question in the field is to understand and control the self-organization of active assemblies in space and time. Most active systems exhibit either spatial order mediated by interactions that coordinate the spatial structure and the motion of active agents or the temporal synchronization of individual oscillatory dynamics. The simultaneous control of spatial and temporal organization is more challenging and generally requires complex interactions, such as reaction-diffusion hierarchies or genetically engineered cellular circuits. Here, we report a novel and simple means to simultaneously control the spatial and temporal self-organization of bacterial active matter. By confining an active bacterial suspension and manipulating a single macroscopic parameter, namely the viscoelasticity of the suspending fluid, we have found that the bacterial fluid first self-organizes in space into a millimeter-scale rotating vortex; then displays temporal organization as the giant vortex switches its global chirality periodically with tunable frequency, reminiscent of a torsional pendulum - a self-driven one. Combining experiments with an active matter model, we explain this striking behavior in terms of the interplay between active forcing and viscoelastic stress relaxation. Our findings advance the understanding of bacterial behavior in complex fluids, and demonstrate experimentally for the first time that rheological properties can be harnessed to control active matter flows. Coupled with actuation, our tunable self-oscillating bacterial vortex may be used as a "clock" for locomotion of soft robots and microfluidic pumping.

cond-mat.soft↗