arXiv subjects

Haochen Shi

Publications and source records attributed to Haochen Shi.

At least 19 recordsLinked to original sources

arXiv2026

Finding the Right Evidence: Factor-Guided Coarse-to-Fine Reasoning for Long Videos

While LVLMs rapidly improve, long-video question answering still remains challenging: relevant evidence is sparse, and question-relevant context often fails to provide cues that discriminate the correct answer from plausible alternatives. Diagnostic analysis on a manually annotated subset of MMR-V shows that prior agentic systems substantially improve cue retrieval over direct VLM inference yet fail to achieve a corresponding gain in answer accuracy, indicating that the bottleneck lies in option-discriminative evidence rather than topical relevance alone. We propose PACE (Progressive Acquisition of Critical Evidence), a factor-guided framework for long-video evidence acquisition. PACE proceeds in two stages: it first indexes clip-level descriptions guided by question-derived factors without observing the candidate answers; it then uses the candidate answers to derive contrastive cues and queries the index for verification. On MMR-V with the open-source Qwen3-VL backbone, PACE achieves 42.6% accuracy, outperforming direct inference and prior agentic baselines including Deep Video Discovery (DVD). On the same diagnostic subset, PACE recovers 66.9% of the annotated cues, providing empirical evidence that its gains are associated with improved evidence recovery rather than stronger answer-side priors alone. Consistent gains over DVD on LVBench, Video-MME, EgoSchema, and LongVideoBench suggest that option-aware evidence acquisition transfers beyond MMR-V. Code is available at https://github.com/HKUST-KnowComp/PACE.

cs.CV↗

arXiv2026

MultivationBench: A Benchmark for Multimodal Sequential Motivation Reasoning

Multimodal Large Language Models have sparked significant interest due to their potential for social intelligence; however, their ability to perform sequential motivation reasoning remains insufficiently studied. Existing evaluations predominantly examine static text or isolated visual snapshots, which do not reflect the cumulative nature of real-world behavioral drivers. To address this gap, we introduce MultivationBench, a benchmark designed to rigorously evaluate multimodal motivation reasoning within story-driven visual narratives. The benchmark builds upon established psychological frameworks - Maslow's hierarchy and Reiss's basic desires - and requires models to integrate accumulated multimodal context to infer evolving motivations. Results indicate that MultivationBench presents a significant challenge: all tested models struggle to maintain consistent motivation reasoning across sequential contexts, revealing a critical disconnect between static recognition capabilities and the dynamic reasoning essential for human-like social understanding.

cs.AI↗

arXiv2026

RLPF: Reinforcement Learning from Performance Feedback for Code Generation

Code models are increasingly trained with execution feedback, but most training signals still stop at correctness. This leaves an important gap for systems code: two programs can pass the same tests while differing greatly in runtime. We study how to train code agents to prefer faster correct implementations, rather than treating efficiency only as an evaluation metric. The key difficulty is that runtime is a fragile reward. It is meaningful only after a program is correct, varies across tasks, and gives little guidance when most sampled programs fail to compile or run. We propose \textbf{RLPF}, reinforcement learning from performance feedback, which turns execution outcomes into a staged reward. Failed programs are ordered by execution progress, while correct programs are ranked by their relative improvement from the baseline toward the expert reference. This gives useful feedback before correctness and performance-sensitive feedback after correctness. Fine-tuning Qwen3-32B with RLPF on PerfCodeBench raises correct-and-runnable solutions from $11.1\%$ to $54.6\%$ and improves relative efficiency from $8.1\%$ to $38.6\%$. The trained model becomes competitive with stronger open-weight systems, and its optimization behavior transfers modestly to EffiBench-X. Additional studies show that model-generated references provide useful but weaker supervision, and that the full composite reward is more reliable than correctness-only or runtime-only baselines. These results suggest that code agents can be trained not only to pass tests, but also to optimize the programs they write.

cs.LG↗

arXiv2026

DishSeg24k: A Large-Scale Benchmark for Food Segmentation with Stochastic Expert Decoding

Food segmentation is essential for applications such as intelligent catering, dietary assessment, and recommendation. However, existing benchmarks fail to capture the complexity of real-world dining scenes. The challenges of dense inter-dish overlap, fine-grained class similarity, and extreme long-tail class distributions exceed the fidelity of current datasets. To fill this gap, we introduce \textbf{DishSeg24k}, a large-scale dish-level segmentation benchmark with 24,096 images, 112,281 instances, and 278 fine-grained categories in real-world dining environments. Based on DishSeg24k, we further propose \textbf{Food Expert-Adaptive Segmentation Transformers (FEAST)} to address these challenges. FEAST models query-based decoding as a Markov Decision Process (MDP), where each decoder layer update is treated as a sequential decision step that explores uncertainty along dish boundaries. We further redesign the decoder with a reinforcement learning (RL)-guided Mixture-of-Experts (MoE) module, in which a dual-critic decoupled optimization scheme separates task-oriented query refinement from structure-aware expert routing. This design promotes expert specialization and prevents expert collapse under long-tail category distributions. Finally, extensive experiments on DishSeg24k demonstrate the state-of-the-art performance of FEAST, which outperforms previous methods by {+3.21\%} mIoU, {+3.68\%} mDice, and {+4.00\%} mAcc, respectively. We further validate the effectiveness of FEAST on FoodSeg103. The dataset and code will be publicly released.

cs.CV↗

arXiv2026

Handroid: Bridging Dexterous Hand and Humanoid

Dexterous hands and humanoid robots are typically developed as distinct embodiments: the former enable contact-rich manipulation at the object scale, whereas the latter provide mobility and whole-body interaction in human-centered environments. We introduce \textbf{Handroid}, a desktop-scale dual-embodiment robot that integrates both capabilities within a single reconfigurable platform. Handroid reuses one 27-DoF electromechanical body as either a dexterous hand or a desktop humanoid, measuring 0.33 m in height and 2.05 kg in weight. In the dexterous hand embodiment, 20 DoFs form an anthropomorphic hand closely matching the kinematic structure of the human hand. In the humanoid embodiment, the same articulated modules are reconfigured into a humanoid with a head, arms, and legs, including a 12-DoF lower-limb structure for locomotion and whole-body motion. Handroid further provides a unified control and learning framework supporting hand teleoperation, dexterous grasping, in-hand manipulation, humanoid locomotion, gait generation, and interactive motion authoring. We validate the platform through real-world dexterous manipulation, reinforcement-learning-based locomotion, keyframe motion deployment, and a long-horizon task involving embodiment reconfiguration, locomotion, docking, and dexterous pick-and-place. These results position Handroid as a compact and reproducible platform for advancing morphology-reconfigurable robotics and cross-embodiment robot learning.

cs.RO↗

arXiv2026

Isolation as a First-Class Principle for LLM-Agent System Safety: Concepts, Taxonomy, Challenges and Future Directions

The capability of LLM agents to function as the ``brain'' of a system fundamentally expands the scope of analysis beyond a standalone model. Consequently, safety is no longer only about input--output content alignment. It also concerns system behavior and real-world execution outcomes. However, the current literature is fragmented across attack types, applications, and benchmarks. This makes it hard to explain why failures such as prompt injection, tool misuse, and memory poisoning often share the same structural cause, and how they spread through an agent workflow. In this survey, we treat isolation as a first-class principle for LLM-agent system safety. By isolation, we refer to the separation of user inputs, tool access, execution channels, inter-agent communication, and environment-originated context. We organize the literature with a boundary-centric taxonomy of five boundaries: user-agent, agent-tool, agent-execution, agent-agent, and system-environment. This view helps identify where the loss of isolation first occurs, how compromise propagates across boundaries, and which defenses are most relevant at each interface. We also summarize cross-boundary failure paths, discuss open challenges, and outline a research agenda for isolation-by-construction in future agent systems.

cs.AI↗

arXiv2026

PersonalPlan: Planning Multi-Agent Systems for Personalized Programming Learning

Effective programming education requires personalized instruction adapted to diverse learner backgrounds. However, while LLM-based multi-agent systems (MAS) excel at complex planning, existing planners often lack profile-grounding and pedagogical scaffolding, thereby undermining personalized programming learning. To fill in the gap, we first introduce \textbf{MAP-PPL} (\textbf{M}ulti-\textbf{A}gent \textbf{P}lans for \textbf{P}ersonalized \textbf{P}rogramming \textbf{L}earning), a profile-conditioned multi-agent planning dataset with 3{,}043 query--profile--plan instances from 1{,}730 Stack Overflow question groups and 2{,}738 learner profiles. Each plan specifies agents, subtasks, executable steps, and prerequisite dependencies. Then, we propose \textbf{PersonalPlan}, a two-stage MAS planner that first performs hierarchical SFT with separate LoRA adapters for profile-aware task decomposition and step dependency planning, then applies a Reward-Adaptive GRPO to encourage the model to generate executable, personalized, and pedagogically scaffolded plans. Extensive experiments on MAP-PPL comparing PersonalPlan against frontier LLMs, generic MAS frameworks, and agentic planners demonstrate its superiority. With only 8B and 32B variants, PersonalPlan achieves state-of-the-art plan executability, personalization, and pedagogical quality, effectively orchestrating MAS for agent-student interactions.

cs.MA↗

arXiv2026

SING: Synthetic Intention Graph for Scalable Active Tool Discovery in LLM Agents

Large language model (LLM) agents increasingly rely on agent harnesses that manage context, tools, and multi-turn execution, making tools a central interface for acting in realistic digital environments. As harness-connected tool ecosystems expand to hundreds or thousands of APIs, services, and task-specific skills, exhaustive tool schema injection becomes costly and imposes a closed-world assumption that limits agents to a predefined static inventory. Retrieval-augmented tool selection offers a natural alternative, but existing one-shot retrieval methods often fail to align isolated tool descriptions with the agent's true task intention, especially in long-horizon tasks where required capabilities emerge through decomposition, observations, and newly induced subgoals. We propose SING, an intention-aware active tool discovery framework that builds an intention-tool graph linking user intentions, tool capabilities, and tool collaboration patterns, and dynamically retrieves tools according to evolving task states. Using a unified corpus of 7,471 tools, we evaluate SING on three real-world tool-use benchmarks. SING improves Global Recall@5 by up to 59.8% and downstream success rate by up to 28.9% over baselines, while reducing full-corpus tool-schema exposure by 99.8%, demonstrating that intention-aware graph structure enables more accurate and context-efficient tool discovery in large-scale agentic ecosystems.

cs.CL↗

arXiv2026

PerfCodeBench: Benchmarking LLMs for System-Level High-Performance Code Optimization

Large language models (LLMs) can often generate functionally correct code, but their ability to produce efficient implementations for performance-critical systems tasks remains limited. Existing code benchmarks mainly emphasize correctness or algorithmic problem solving, while realistic systems-level optimization is still underexplored. To address this gap, we introduce PerfCodeBench, an executable benchmark for evaluating LLMs on high-performance code optimization. The tasks require system-level implementation choices, hardware-aware optimization, and careful handling of performance bottlenecks. Each task includes executable correctness checks, a baseline implementation, and a reference optimized solution. This allows us to evaluate both correctness and runtime-oriented efficiency. Our evaluation on a broad set of state-of-the-art LLMs shows a clear gap between model-generated code and expert-optimized implementations. The gap is especially large on tasks involving parallelism and GPU operations. Current models also show weaknesses in cross-language robustness and in consistently reaching expert-level efficiency. These results suggest that performance-aware evaluation are still needed. LLMs should move beyond generating merely correct code toward producing efficient systems software. We submit the benchmark data, evaluation infrastructure, and complete logs of all LLMs-generated code at https://anonymous.4open.science/r/perfcodebench-7CDE.

cs.SE↗

arXiv2026

OmniCompliance-100K: A Multi-Domain, Rule-Grounded, Real-World Safety Compliance Dataset

Ensuring the safety and compliance of large language models (LLMs) is of paramount importance. However, existing LLM safety datasets often rely on ad-hoc taxonomies for data generation and suffer from a significant shortage of rule-grounded, real-world cases that are essential for robustly protecting LLMs. In this work, we address this critical gap by constructing a comprehensive safety dataset from a compliance perspective. Using a powerful web-searching agent, we collect a rule-grounded, real-world case dataset OmniCompliance-100K, sourced from multi-domain authoritative references. The dataset spans 74 regulations and policies across a wide range of domains, including security and privacy regulations, content safety and user data privacy policies from leading AI companies and social media platforms, financial security requirements, medical device risk management standards, educational integrity guidelines, and protections of fundamental human rights. In total, our dataset contains 12,985 distinct rules and 106,009 associated real-world compliance cases. Our analysis confirms a strong alignment between the rules and their corresponding cases. We further conduct extensive benchmarking experiments to evaluate the safety and compliance capabilities of advanced LLMs across different model scales. Our experiments reveal several interesting findings that have great potential to offer valuable insights for future LLM safety research.

cs.CL↗

arXiv2026

TRACE: Temporal Rule-Anchored Chain-of-Evidence on Knowledge Graphs for Interpretable Stock Movement Prediction

We present a Temporal Rule-Anchored Chain-of-Evidence (TRACE) on knowledge graphs for interpretable stock movement prediction that unifies symbolic relational priors, dynamic graph exploration, and LLM-guided decision making in a single end-to-end pipeline. The approach performs rule-guided multi-hop exploration restricted to admissible relation sequences, grounds candidate reasoning chains in contemporaneous news, and aggregates fully grounded evidence into auditable \texttt{UP}/\texttt{DOWN} verdicts with human-readable paths connecting text and structure. On an S\&P~500 benchmark, the method achieves 55.1\% accuracy, 55.7\% precision, 71.5\% recall, and 60.8\% F1, surpassing strong baselines and improving recall and F1 over the best graph baseline under identical evaluation. The gains stem from (i) rule-guided exploration that focuses search on economically meaningful motifs rather than arbitrary walks, and (ii) text-grounded consolidation that selectively aggregates high-confidence, fully grounded hypotheses instead of uniformly pooling weak signals. Together, these choices yield higher sensitivity without sacrificing selectivity, delivering predictive lift with faithful, auditably interpretable explanations.

cs.CE↗

arXiv2026

Minimalist Compliance Control

Compliance control is essential for safe physical interaction, yet its adoption is limited by hardware requirements such as force torque sensors. While recent reinforcement learning approaches aim to bypass these constraints, they often suffer from sim-to-real gaps, lack safety guarantees, and add system complexity. We propose Minimalist Compliance Control, which enables compliant behavior using only motor current or voltage signals readily available in modern servos and quasi-direct-drive motors, without force sensors, current control, or learning. External wrenches are estimated from actuator signals and Jacobians and incorporated into a task-space admittance controller, preserving sufficient force measurement accuracy for stable and responsive compliance control. Our method is embodiment-agnostic and plug-and-play with diverse high-level planners. We validate our approach on a robot arm, a dexterous hand, and two humanoid robots across multiple contact-rich tasks, using vision-language models, imitation learning, and model-based planning. The results demonstrate robust, safe, and compliant interaction across embodiments and planning paradigms.

cs.RO↗

arXiv2026

SurveyLens: A Discipline-Aware Benchmark for Automatic Survey Generation

Automatic Survey Generation (ASG) aims to produce comprehensive literature surveys by retrieving, organizing, and synthesizing academic papers. Despite rapid progress in specialized ASG frameworks and Deep Research agents, existing evaluations largely center on Computer Science or rely on generic criteria, leaving it unclear whether current systems satisfy the survey standards of diverse disciplines. We introduce SurveyLens, the first discipline-aware ASG benchmark. SurveyLens comprises SurveyLens-1k, a curated dataset of 1,000 human-written surveys across 10 disciplines, and a dual-lens framework that combines discipline-aware rubric scoring with reference-based alignment to human-written surveys. Evaluating 11 state-of-the-art systems across vanilla LLMs, ASG systems, and Deep Research agents, we find that Deep Research agents are the only paradigm robust across all 10 disciplines, ASG systems lead on structural planning, and all paradigms remain weak on reference quality, providing practical guidance for discipline-specific tool selection and future ASG design.

cs.CL↗

arXiv2026

History-Guided Iterative Visual Reasoning with Self-Correction

Self-consistency methods are the core technique for improving the reasoning reliability of multimodal large language models (MLLMs). By generating multiple reasoning results through repeated sampling and selecting the best answer via voting, they play an important role in cross-modal tasks. However, most existing self-consistency methods are limited to a fixed ``repeated sampling and voting'' paradigm and do not reuse historical reasoning information. As a result, models struggle to actively correct visual understanding errors and dynamically adjust their reasoning during iteration. Inspired by the human reasoning behavior of repeated verification and dynamic error correction, we propose the H-GIVR framework. During iterative reasoning, the MLLM observes the image multiple times and uses previously generated answers as references for subsequent steps, enabling dynamic correction of errors and improving answer accuracy. We conduct comprehensive experiments on five datasets and three models. The results show that the H-GIVR framework can significantly improve cross-modal reasoning accuracy while maintaining low computational cost. For instance, using \texttt{Llama3.2-vision:11b} on the ScienceQA dataset, the model requires an average of 2.57 responses per question to achieve an accuracy of 78.90\%, representing a 107\% improvement over the baseline.

cs.CL↗

arXiv2026

NOVA: NOise-aware Verbal Confidence CAlibration for Robust Large Language Models in RAG Systems

Accurately assessing model confidence is essential for deploying large language models (LLMs) in mission-critical factual domains. While retrieval-augmented generation (RAG) is widely adopted to improve grounding, confidence calibration in RAG settings remains poorly understood. We conduct a systematic study across four benchmarks, revealing that LLMs exhibit poor calibration performance especially when noisy contexts are retrieved. Specifically, contradictory or irrelevant evidence tends to exacerbate the model's overconfidence issue. To address this, we propose NOVA Rules (NOise-Aware Verbal Confidence CAlibration Rules) to provide a principled foundation for resolving overconfidence under noise. We further design NOVA, a noise-aware calibration framework that synthesizes supervision from ~2K HotpotQA examples guided by these rules. By performing supervised fine-tuning (SFT) with this data, NOVA equips models with intrinsic noise awareness without relying on stronger teacher models. Empirical results show that NOVA yields substantial gains, improving ECE scores by 10.9% in-domain and 8.0% out-of-domain. By bridging the gap between retrieval noise and verbal calibration, NOVA paves the way for both accurate and epistemically reliable LLMs.

cs.CL↗

arXiv2026

Evolving Programmatic Skill Networks

We study continual skill acquisition in open-ended embodied environments where an agent must construct, refine, and reuse an expanding library of executable skills. We introduce the Programmatic Skill Network (PSN), a framework in which skills are executable symbolic programs forming a compositional network that evolves through experience. PSN defines three core mechanisms instantiated via large language models: (1)~\opreflect for structured fault localization over skill compositions, (2)~progressive optimization with maturity-aware update gating that stabilizes reliable skills while maintaining plasticity for uncertain ones, and (3)~canonical structural refactoring under rollback validation that maintains network compactness. We further show that PSN's learning dynamics exhibit structural parallels to neural network training. Experiments on MineDojo and Crafter demonstrate robust skill reuse, rapid adaptation, and strong generalization across open-ended task distributions.

cs.AI↗

arXiv2026

Locomotion Beyond Feet

Most locomotion methods for humanoid robots focus on leg-based gaits, yet natural bipeds frequently rely on hands, knees, and elbows to establish additional contacts for stability and support in complex environments. This paper introduces Locomotion Beyond Feet, a comprehensive system for whole-body humanoid locomotion across extremely challenging terrains, including low-clearance spaces under chairs, knee-high walls, knee-high platforms, and steep ascending and descending stairs. Our approach addresses two key challenges: contact-rich motion planning and generalization across diverse terrains. To this end, we combine physics-grounded keyframe animation with reinforcement learning. Keyframes encode human knowledge of motor skills, are embodiment-specific, and can be readily validated in simulation or on hardware, while reinforcement learning transforms these references into robust, physically accurate motions. We further employ a hierarchical framework consisting of terrain-specific motion-tracking policies, failure recovery mechanisms, and a vision-based skill planner. Real-world experiments demonstrate that Locomotion Beyond Feet achieves robust whole-body locomotion and generalizes across obstacle sizes, obstacle instances, and terrain sequences.

cs.RO↗

arXiv2025

CritiCal: Can Critique Help LLM Uncertainty or Confidence Calibration?

Accurate confidence calibration in Large Language Models (LLMs) is critical for safe use in high-stakes domains, where clear verbalized confidence enhances user trust. Traditional methods that mimic reference confidence expressions often fail to capture the reasoning needed for accurate confidence assessment. We propose natural language critiques as a solution, ideally suited for confidence calibration, as precise gold confidence labels are hard to obtain and often require multiple generations. This paper studies how natural language critiques can enhance verbalized confidence, addressing: (1) What to critique: uncertainty (question-focused) or confidence (answer-specific)? Analysis shows confidence suits multiple-choice tasks, while uncertainty excels in open-ended scenarios. (2) How to critique: self-critique or critique calibration training? We propose Self-Critique, enabling LLMs to critique and optimize their confidence beyond mere accuracy, and CritiCal, a novel Critique Calibration training method that leverages natural language critiques to improve confidence calibration, moving beyond direct numerical optimization. Experiments show that CritiCal significantly outperforms Self-Critique and other competitive baselines, even surpassing its teacher model, GPT-4o, in complex reasoning tasks. CritiCal also shows robust generalization in out-of-distribution settings, advancing LLM's reliability.

cs.CL↗