Searcharxiv⌕ Search

arXiv subjects

Yifan Zhu

Publications and source records attributed to Yifan Zhu.

At least 19 recordsLinked to original sources

Objectives as Policies: Vision-Language Model Cost Programs for Contact-Rich Manipulation

Vision-language models (VLMs) have shown strong semantic and spatial reasoning capabilities in 3D environments. Recently, connecting VLMs with model-predictive controllers (MPC) has emerged as a paradigm for combining strong semantic reasoning with physics-based planning. However, connecting these two for contact-rich robotic manipulation requires a task representation that bridges the two and captures how physical interactions can be reasoned by VLMs. In this work, we introduce Objectives as Policies (OaP), a framework that connects a pretrained VLM to a shared sampling-based MPC through an editable objective program. Each program specifies a sequence of stages for completing a task, combining running and terminal costs from a cost library with stage completion conditions. Given a natural-language task instruction and scene observation, we build a simulation-based digital twin through real-to-sim, and the VLM generates an objective program that is executed by the MPC in the simulation. The program is iteratively updated by the VLM based on MPC rollouts, before being deployed in the real world. We evaluate our framework in both simulation and the real world on four tabletop manipulation tasks, and show that it outperforms strong baselines. Additional videos and examples are available on the project website: https://zihaolu001.github.io/oap-project/

cs.RO↗

LoopWeaver: Weaving Feedback Loops into Hierarchical Generation under Constraints

Large Language Models face a multi-objective optimization problem in constrained long-text generation, balancing global structure, local coherence, and constraint satisfaction. We propose LoopWeaver, a hierarchical workflow with two sequential offline preference-optimization stages. Constraint-aware planning produces structured plans, feasibility filtering removes invalid plans, and a shared deterministic reward constructs preference pairs for plan-level and generation-level DPO. At inference time, LoopWeaver remains feed-forward: it generates and revises a complete plan, filters the revised plan, and then generates segments conditioned on the current sub-plan and running narrative. We evaluate the method on four LongGenBench-style structured scenarios across multiple backbones. Results show improvements in both constraint accuracy and text quality across all four open-source backbones. Our software and data are publicly available at https://github.com/cgtaa-cmyk/LoopWeaver.

cs.CL↗

MMGraph-Agent: Agentic Multimodal RAG via Cache-Inspired Multimodal Knowledge HyperGraphs

Driven by increasingly complex real-world applications, Retrieval-Augmented Generation (RAG) has evolved from textual pipelines to multimodal settings, comprising two complementary dimensions: knowledge organization and retrieval enhancement. However, existing methods still face challenges of high construction cost, context window limitations, and the lack of unified training across offline and online knowledge sources. To address these challenges, we propose MMGraph-Agent, a unified multimodal RAG framework based on cache-inspired multimodal knowledge hypergraphs. MMGraph-Agent enables extremely lightweight hypergraph construction with zero API cost and dynamic memory-based retrieval that alleviates long-context issues. Experiments across offline, online, and joint retrieval settings demonstrate consistent improvements in performance, efficiency, and architectural unification. Our software and data are publicly available at https://github.com/LHRLAB/MMGraph-Agent.

cs.AI↗

VOMMI: Collecting and Leveraging Portable Demonstrations for Mobile Manipulation

Portable mobile-manipulation demonstrations can help alleviate data scarcity for embodied intelligence, but obtaining reliable, low-cost, and robot-free motion supervision from RGB observations remains challenging. Existing approaches often rely on teleoperation or specialized devices equipped with additional sensing hardware, while directly using estimated visual odometry (VO) trajectories can introduce inconsistencies due to accumulated drift and imperfect motion supervision. We present the Visual-Odometry-Conditioned Mobile Manipulation Interface (VOMMI), a portable demonstration collection and learning framework that connects portable RGB demonstrations to vision-language-action (VLA) post-training through offline trajectory reconstruction and online visual-motion conditioning. VOMMI synchronizes body and hand views to capture navigation context and local object interactions without requiring human-robot kinematic correspondence calibration. R2-VO refines offline demonstration trajectories using sparse geometric anchors and produces causal local-motion tokens over multiple prediction horizons for online policy conditioning. An action-group residual adapter incorporates these tokens only into the base branch. Experiments use a 500-trajectory portable for each task, with 75 trajectories held out for RGB-VO evaluation, and 200 robot demonstrations as references. Our policy, post-trained only on portable demonstrations, achieves 18.2% lower base-velocity error than a policy trained with robot-collected demonstrations, while maintaining comparable end-effector translation accuracy. Offline reconstruction reduces absolute trajectory errors for the body and hand streams by 24.6% on average relative to the best evaluated baseline for each stream. The complete system improves the mean success rate by 8.3 percentage points over OpenPI 0.5 across three real-robot tasks.

cs.RO↗

SQL-o1: Automatic Text-to-SQL Program Synthesis via Self-Reward Heuristic Search

Efficient human-database interaction is a core problem in software engineering. Text-to-SQL (Text2SQL) tackles this by mapping natural language questions to executable SQL queries, enabling automatic SQL synthesis and reliable interaction. Although large language models (LLMs) have advanced significantly, current approaches struggle with poor transferability to open-source LLMs, limited robustness against logic and function errors in complex queries, and inefficiencies in structured search. We introduce SQL-o1, a self-reward-driven heuristic search framework built on an agent-based architecture to enhance model reasoning capabilities. SQL-o1 leverages Monte Carlo Tree Search (MCTS) for structured, multi-step exploration and incorporates a dynamic pruning strategy to accelerate inference without sacrificing accuracy. On the Spider and BIRD benchmarks, SQL-o1 achieves a 10.8-point execution accuracy improvement on the complex BIRD dataset, surpassing even GPT-4-based models. It exhibits strong few-shot generalization and robust cross-model transferability across open-source LLMs. Our code is available at https://github.com/ShuaiLyu0110/SQL-o1.

cs.DB↗

GUI-AC: Enhancing Continual Learning in GUI Agents

Graphical User Interfaces (GUIs) serve as the dominant medium for human-computer interaction, yet building GUI agents that generalize across the vast diversity of real-world interface environments, with the same flexibility and robustness that humans naturally exhibit, remains unsolved. Notably, GUI data are inherently non-stationary: the continual emergence of previously unseen interface instances (e.g., novel domains and resolutions) induces persistent distribution shifts, significantly impeding the continual learning of existing GUI agents. Reinforcement fine-tuning (RFT) has attracted considerable attention as a promising approach. Nevertheless, RFT exhibits pronounced instability in its grounding capability, where rollout outcomes are highly imbalanced and fixed clipping can suppress valuable exploration. More importantly, policy probability alone does not reveal the semantic role of a grounding prediction. To address these challenges, we propose GUI-AC, a continual RFT framework that jointly characterizes grounding rollouts through advantage and policy probability. This yields a four-region view of grounding behavior and enables targeted optimization of Exploration and Overconfidence. GUI-AC further introduces grounding certainty to support two core mechanisms: (i) Adaptive Advantage, which strengthens the negative correction for overconfident grounding failures; and (ii) Dynamic Clipping, which relaxes the upper clipping bound for exploratory grounding rollouts. Extensive experiments show that GUI-AC consistently outperforms state-of-the-art baselines. Code is available at https://anonymous.4open.science/r/GUI-AC-Code.

cs.CV↗

AI Cosmos: An Agentic Infra for Situated Learning

Large language models (LLMs) can improve task performance during learning, yet these gains do not necessarily reflect learners' ability to thereafter apply the knowledge and skills independently. We present AI Cosmos, an agentic infrastructure for situated learning that organizes learning around two complementary principles: scenario-based participation and cumulative assets of practice. AI Cosmos organizes learners and role-specialized agents within learning scenarios that integrate task-relevant knowledge and, when needed, executable workbenches. An asset-memory loop preserves the processes and outcomes of practice as reusable assets with explicit provenance and dependencies. We evaluate scenario-based interaction through two preregistered randomized controlled studies examining the acquisition of legal reasoning and programming competence, holding the same underlying LLM across conditions. AI Cosmos achieves greater learning gains than a conventional LLM in legal reasoning and higher mastery of unseen programming problems than an LLM-as-Teacher baseline.

cs.CL↗

Self-Labeled Preference Transfer for Steerable Robot Manipulation

Vision-language-action (VLA) policies can learn to follow execution preferences, such as where to grasp an object or how to move around an obstacle, from demonstrations annotated with preference labels. Providing these annotations for every new task, however, is costly. We study whether learned preference control can transfer to new tasks whose demonstrations contain the desired behaviors but lack preference annotations. We find that a policy trained with preference annotations can follow preferences on a new task before adaptation but achieves limited task success. Fine-tuning on target demonstrations without these annotations improves task success while weakening preference control. Motivated by this finding, we introduce Self-Labeled Preference Transfer (SPT). SPT jointly trains a shared vision-language backbone for preference-conditioned action prediction and preference recognition on annotated source demonstrations. It then freezes the recognizer to preserve its cross-task preference recognition capability and uses it to self-label target demonstrations for policy fine-tuning. Across five types of execution preferences in simulation, the recognizer achieves 96.4% accuracy, and SPT reaches 84.8% preference following and 89.2% task success, compared with 59.4% and 88.0% for naive finetuning. We further extend to multiple preferences, achieving 93.8% preference following compared with 42.5% of baselines. In real-world experiments, SPT increases preference following from 52.5% to 92.5% relative to naive finetuning across two task transfers. These results show that SPT can support preference adaptation across tasks and objects while maintaining success, without additional manual preference annotations.

cs.RO↗

Same Evidence, Different Answers: Canonical-Context On-Policy Distillation for Multi-Turn Language Models

Large language models (LLMs) often solve a task when all instructions are given in a single prompt, but fail when the same information is revealed gradually across turns. When a clean FULL prompt and a RAW-SHARDED conversation contain the same complete user evidence, the model should still arrive at the same answer. We argue that a key reason for this gap is self-anchored drift: responses produced under partial information introduce unsupported assumptions, and those assumptions later distort the final answer. To reduce this effect, we propose Canonical-Context On-Policy Distillation (CCOPD). During training, the same base model is used in two roles: a frozen teacher conditioned on the clean FULL prompt and a trainable student that receives the same evidence incrementally through a multi-turn conversation; CCOPD aligns the student's behavior on its own trajectories with the teacher's canonical full-context behavior. Trained only on math problem conversations, CCOPD yields a 32\% average relative improvement in RAW-SHARDED performance over the original base model across math and five zero-shot out-of-domain task families, while largely preserving full-context performance. Further analyses suggest that CCOPD strengthens grounding in user evidence and reduces sensitivity to contamination from earlier assistant turns.

cs.CL↗

OmniConfess: Eliciting Token Confessions to Mitigate Omni-Modal Hallucination

Omni-modal large language models (OmniLLMs) unify text, images, audio, and video, yet hallucinate when generation relies on the wrong evidence. Existing inference-time methods can reduce hallucinations, but rarely reveal which evidence sustains a generated commitment. We introduce OmniConfess, a training-free method for mitigating omni-modal hallucinations. It fixes a candidate response and re-scores it at token resolution under controlled channel-wise evidence interventions, producing a structured token-by-channel confession that reveals the response's evidential dependence. OmniConfess uses this confession to preserve grounded content and correct commitments driven by irrelevant or contradictory evidence. To evaluate OmniConfess, we construct OmniHalluBench, a 3,540-example benchmark built from six datasets spanning text, image, audio, and video settings and both judgment and free-form generation. Experiments show that OmniConfess mitigates hallucinations across heterogeneous modality and task settings. Our code and benchmark are publicly available at https://github.com/RongHuiQiang/OmniConfess.

cs.CL↗

InternW0: A Foundational Physical World Model for Efficient Real-World Interactions

Physical intelligence requires more than predicting how the world may evolve: predictions must remain actionable as the world continues to change. We introduce InternW0, the first instantiation of the InternW physical world model series from Shanghai AI Laboratory, built around omnimodal interfaces, asynchronous multi-frequency processing, and local physical modeling under partial observations and external influences. InternW0 jointly learns future visual dynamics and continuous robot control through an asymmetric video--action architecture with flow matching. A high-capacity video expert provides longer-horizon predictive context, while a lightweight action expert operates at a faster timescale. Instead of regenerating the future for every action update, InternW0 reuses layerwise K/V and adapts it to newly observed states through observation-conditioned context routing. Domain-specific interfaces and soft prompts support heterogeneous embodiments, while contact-aware post-training incorporates force and tactile signals for contact-rich manipulation. We train InternW0 on approximately 7,200 hours of heterogeneous robot and egocentric data, including EgoLab, a 275-hour real-laboratory egocentric dataset. Evaluation spans simulation benchmarks and real-world scientific tasks, including a 15-stage metal--organic framework synthesis workflow and 5-stage contact- and force-aware dexterous manipulation for general-purpose quantitative pipetting. These results advance scalable, asynchronous, and science-native physical world models for universal and efficient real-world interactions.

cs.RO↗

Mind the Gap: Theory-of-Mind-Grounded Friction for Epistemic Alignment

Productive dialogue alignment requires distinguishing \emph{surface coordination} (acknowledgments and smooth task progression) from \emph{epistemic alignment} (convergence of belief states); standard preference-based methods typically optimize response-level preferences without explicitly modeling the latter. We operationalize Theory-of-Mind (ToM) inference as a control signal within Frictive Policy Optimization by extracting, at each referring expression, a four-part belief structure: the speaker's intended referent, the addressee's interpretation, and each participant's model of the other's belief. This makes friction mechanically computable from epistemic-state comparisons, capturing \emph{silent divergence}, where both participants proceed confidently while grounding to different referents. We evaluate the signal at two levels. At the representation level, ablating the second-order channel reduces misunderstanding recall from $65\%$ to $26\%$. At the policy level, reward-shaping (FAR) and trust-region (FTR) variants improve intervention F1 and warranted-context calibration over DPO, with Brier scores independently supporting the calibration gains. Across three training runs, FAR and FTR remain substantially more stable, whereas DPO varies widely and can degrade intervention competence already present in the base policy. Thus, ToM-grounded friction provides a trainable signal for context-sensitive intervention under referential belief divergence.

cs.CL↗

Diamond Agent: Agentic Control of Federated HPC Resources as a Service

Efficiently aggregating and orchestrating computing power across heterogeneous clusters for HPC workflows faces four practical challenges: preserving workflow context across independently administered clusters, moving large datasets between sites, reasoning about site-specific environments and scheduler policies, and exploiting live queue and resource states for efficient task scheduling. To this end, we design Diamond Agent, an agentic system that enables intelligent execution of HPC workflows across heterogeneous clusters with typed skills as the interface. Diamond Agent provides an agent-facing workspace and skills that unify cross-site resource discovery, resource specification, data movement, task execution, and result retrieval. A centralized Diamond Agent instance can operate multiple supercomputers without being deployed separately on each login node. Diamond Agent translates high-level agent actions into valid site-specific executions, moves data through Globus Transfer, and uses live system capability and queue information to select feasible placements. Its event-driven continuation mechanism decouples agent actions from long-running batch jobs: persistent services monitor remote execution and resume the agent only when a result or decision-relevant event is available. We experiment with 27 hours of telemetry and 19 matched multi-site submission rounds comprising 83 jobs across four production supercomputers. Compared with a fixed-site baseline, Diamond Agent reduces the median additional completion time relative to the fastest observed placement from 42 seconds to 4 seconds, a 10.5x reduction.

cs.DC↗

Finite-sample nonparametric mean tests: Leave-one-out duality and asymptotic optimality

We study finite-sample valid tests of the one-sided mean hypothesis $H_0:μ\leq 1$ against $H_1:μ>1$ for nonnegative random variables. To do so, we develop a leave-one-out dual certificate framework, where certain pointwise inequalities imply p-value validity under the conditional mean null $\mathbb{E}[X_i\mid\mathbf{X}_{-i}]\leq 1$, and which also gives conditions that allow combining dual certificates for p-values to show that their pointwise minimum is also a valid p-value. The framework proves finite-sample validity of Wang and Zhao's nonparametric likelihood-ratio statistic $T_{\mathrm{nplr}}$, yields a new p-value $T_{\mathrm{bin}+}$ extending the Clopper--Pearson binomial test to general nonnegative random variables, and shows that the pointwise minimum $\min\{T_{\mathrm{nplr}},T_{\mathrm{bin}+}\}$ is itself a valid and more powerful p-value. We establish sharp optimality results for such testing problems in two regimes: both $T_{\mathrm{nplr}}$ and $T_{\mathrm{bin}+}$ attain a universal detectability boundary for the null $H_0$ without moment or tail assumptions, and $T_{\mathrm{bin}+}$ attains a nonparametric power lower bound under $n^{-1/2}$-local alternatives to $H_0$. Efficient algorithms and numerical experiments demonstrate substantial finite-sample power gains over existing valid methods.

math.ST↗

Propose to Learn, Learn to Propose: Evaluability-Aware Assistance under Bounded Rationality

AI assistants often collaborate by proposing candidate edits, plans, or designs that users evaluate before adoption. Existing assistance methods focus on proposal quality or user-goal inference, often assuming that the user can reliably evaluate any proposal, which can fail in practice because of bounded rationality. We study evaluability-aware proposal planning, where proposals serve both as task interventions and as probes for learning latent preferences and evaluation constraints, where the resulting belief updates then guide later proposals. We formalise this setting as ProSE, a hidden-parameter sequential assistance problem, and instantiate it with a KL-regularised bounded-rational binary response model in which acceptance trades off value gain against a distance-dependent evaluability penalty. Analysing the planning consequence of this likelihood reveals that likely accepted proposals and informative probes need not coincide, which explains why planners that only pursue acceptance systematically underperform. We operationalise ProSE with \textsc{ProSE-Plan}, a depth-2 Bayes-adaptive planner that scores proposals by possible responses and response-induced posterior beliefs. In controlled graph simulations, \textsc{ProSE-Plan} improves over evaluability-unaware and myopic baselines when evaluation cost is the bottleneck, and a probe-commit ablation confirms that our approach selects informative proposals that simpler methods miss. Our results thus identify user evaluability as a planning-relevant dimension of AI assistance, complementary to generation quality and preference inference.

cs.AI↗

Assessing Parameter Redundancy in Transformers for Jet Tagging

Transformer-based jet taggers, such as the Particle Transformer (ParT) and the More-Interaction Particle Transformer (MIParT), achieve excellent discrimination by exploiting correlations among jet constituents, but often require more trainable parameters than earlier deep-learning taggers. In this paper, we investigate whether comparable discriminating power can be achieved with substantially fewer parameters. We introduce an hourglass structure that replaces the feed-forward networks (FFNs) in the attention blocks while leaving the particle-interaction attention unchanged. We also introduce a lightweight particle-embedding layer to replace the original dense embedding network. Applying both modifications to ParT and MIParT yields the hourglass (HG) variants ParT-HG and MIParT-HG, respectively. We evaluate both models on benchmark datasets for top tagging and quark-gluon discrimination. Both variants retain comparable tagging performance, including background rejection at fixed signal efficiencies, while using only approximately 48% and 39.7% of the parameters of their respective baselines. On the larger JetClass dataset, accuracy and AUC decrease by less than 1%, and background rejection also decreases for several signal classes. Overall, our approach provides an alternative way to reduce the parameter count of Transformer jet taggers while largely retaining their tagging performance.

hep-ph↗

Global Convergence of an SQP Method for Contact-Implicit Trajectory Optimization

Contact-Implicit Trajectory Optimization (CITO) is a powerful framework for planning motions of robots that interact with complex environments, but its convergence behavior remains difficult to characterize. Existing formulations either rely on off-the-shelf nonlinear programming solvers whose guarantees require constraint qualifications that are hard to verify for contact-rich systems, or require differentiable explicit dynamics maps that are difficult to formulate in the presence of impacts, changing contact modes, and geometric non-smoothness. This paper studies CITO for a broader class of constraint-rich dynamic systems formulated through implicit physics constraints. By exploiting maximal-coordinate structure, penalty relaxations of equality and inequality constraints, and finite-horizon Hamiltonian bounds, we prove global convergence in the numerical-optimization sense for a specific line-search Sequential Quadratic Programming (SQP) method with adaptive timestep refinement. Under the stated constraint-rich dynamic-system assumptions and fixed pre-horizon data satisfying the strict-interior predecessor condition, the method terminates finitely from any finite discretized optimized trajectory guess at an $ε$-feasible, force-parameterized stationary point---a unit-objective, penalty-induced force-parameterized Fritz--John certificate for smooth damping, weakening to a homogeneous force-parameterized Clarke--Fritz--John certificate whose objective multiplier may vanish under nonsmooth frictional contact.

math.OC↗

Multi-Branch Policy Optimization for Multimodal Large Language Models

Group-based reinforcement learning methods for multimodal large language models typically rely on trajectory-level credit assignment that applies a single advantage to all tokens in a response. However, multimodal reasoning involves substantially higher perceptual uncertainty than text-only settings, where the model must repeatedly re-examine visual information to verify intermediate interpretations, and different visual groundings can lead to divergent reasoning paths, making such uniform credit assignment particularly inadequate and causing relative advantages to progressively degenerate toward zero. To address these challenges, we propose Multi-Branch Policy Optimization (MBPO), a tree-based framework that constructs reasoning trees at vision-language decision boundaries, enabling sibling branches to explore diverse visual hypotheses and assigning segment-level credit through branch-relative advantages. We further introduce a temporal replay buffer to reuse informative segments while controlling policy staleness. Experiments on several multimodal reasoning benchmarks show that MBPO outperforms representative baselines, improving both learning signal quality and optimization efficiency. The code is publicly available at https://github.com/ShuaiLyu0110/MBPO.

cs.CV↗