arXiv subjects

Peng Qi

Publications and source records attributed to Peng Qi.

At least 19 recordsLinked to original sources

arXiv2026

Towards Trustworthy Physical Intelligence: From Theory to Practice Across Life Cycle

Physical intelligence refers to intelligence systems that understand, reason about and act in accordance with the physical world and its underlying laws, dynamics, and constraints. Unlike conventional AI systems, physical intelligence interacts continuously with uncertain physical environments, and its actions produce consequences that are physically irreversible. As existing trustworthy AI frameworks have been developed primarily for digital AI systems, they do not fully capture the distinctive challenges of Physical Intelligence, such as physical safety, cyber-physical security, and physical manufacturing process. To address this gap, our paper develop a foundation for understanding, operationalizing, and governing trustworthy physical intelligence. First, we characterize the fundamental capabilities of physical intelligence. Second, we investigate the role of physical laws and principles in AI and analyze the associated challenges of physical intelligence. Third, we provide a unified structure of the end-to-end physical intelligence life cycle across five core stages and introduce the Trustworthy Physical Intelligence Operationalization (T-PAIO) framework. Fourth, we formulate the Trustworthy Physical Intelligence (T-PAI) Principles as a theoretical framework that synthesizes key trustworthiness dimensions and establishes a conceptual foundation for the governance of physical intelligence systems.

cs.AI↗

arXiv2026

Recoverability and Verifier Strength: Two Measured Preconditions for Unattended Agent Self-Improvement

Agent platforms are starting to improve their own agents: they read a deployed agent's execution traces, derive a change, have an automated verifier approve it, and release it with no human author. How much of this loop can safely run without a human? We propose two quantities a platform can measure. Verifier strength $σ$ is how well the available verifier -- often an LLM judge -- agrees with a trusted reference verifier; it limits how much a step can improve the agent. Recoverability is how cheaply a bad change can be undone: whether undoing it needs a new verifier approval, and how much of the old behavior a rollback restores. A step should run unattended only if both are adequate. On our platform this rule allows prompt, tool and skill edits to run unattended and keeps weight updates under human sign-off. Our main contribution is these two definitions and how to measure them. We built a platform that applies the rule and tested it in two ways: on the platform's own approval gate, and in controlled experiments on the $τ^2$-bench agent benchmark. On verifier strength, the platform's LLM judge sometimes gave the same score no matter what it was shown (9 of 37 settings) and reported no error, so a verifier can look healthy while approving anything; its strength must be measured, and two simple tests catch the failure. On recoverability, restoring a saved copy brought the agent back to its original behavior, but a regenerated undo did not, so only changes whose old version is kept can be reversed without a human; and while a change was deployed, it altered the outcome of 17 percent of tasks beyond normal run-to-run variation, which no undo repairs. Together these support the rule: a change can run unattended only when its old version is kept and its verifier's strength has been measured.

cs.AI↗

arXiv2026

One Task, Many Interfaces: Simple Cross-Interface Consistency Training Improves LLM Agent Robustness

Post-training binds the competence of an LLM agent to the interface it was trained on. Semantics-preserving perturbations of that interface, such as renaming an action or changing the notation, leave the underlying task unchanged yet make the agent worse at it. Scoring each sample independently means interface randomization only broadens this dependence from a single interface to the training interface distribution, so out-of-distribution perturbations still degrade the agent. We propose Cross-Interface Policy Regularization (CIPR), a self-supervised objective that replaces action names with uninformative identifiers and presents each training sample under two interfaces in which equivalent actions are identified by construction. CIPR penalizes the KL divergence between the action distributions induced under the two interfaces, so the policy behaves consistently across them. We evaluate CIPR in both supervised and reinforcement-learning post-training, where it improves robustness. On agentic function calling, held-out interface perturbations reduce the accuracy of a post-trained agent by nearly 25%, and CIPR recovers 10.5% while leaving accuracy on the unchanged interface intact.

cs.LG↗

arXiv2026

Agent Behavior as Code: Efficient and Robust LLM Agents with Programmatic Specifications

AI agents based on foundation models (FMs) have demonstrated strong capabilities to perform complex open-ended tasks. However, they face some common challenges in practice: (a) agent behavior can deviate drastically even for semantically similar tasks, leading to catastrophically propagated errors; (b) high cost and latency due to FM calls, repeated in full whenever a task recurs with different inputs; (c) FMs' limited context and instruction following capability confine how well agents manage the ever-growing execution context and follow complex plans. We introduce $\textbf{A}$gent $\textbf{B}$ehavior as $\textbf{C}$ode $\textbf{Agent}$ (ABCAgent), which uses a symbolic program (e.g., Python code with potential neural functions) to fully specify the agent's behavior at runtime, with a powerful FM agent editing that program for flexibility. Behavior is thus specified without premature variable binding, and its execution is deterministic. We evaluate ABCAgent on six agent benchmarks, two of which we construct to test how well a derived program generalizes to variants of the task it was written for. ABCAgent matches a model-matched neural agent on GAIA and augmented GAIA, and surpasses it where robustness and long control flows matter: 98.3% against 97.3% on GSM-Symbolic ($p = 0.001$), 71.9% against 47.4% $\mathrm{Pass}^4$ on the telecom domain of $τ^2$-bench ($p = 0.0001$), and more records written correctly at every loop length on our control-flow-augmented WorkArena benchmark. For more parametric task families, ABCAgent is also significantly superior in efficiency. Without authoring a new program, ABCAgent solves 92.6% of GSM-Symbolic instances and 20.1% of augmented GAIA variants, which yields $5.2\times$ lower latency and $7.0\times$ lower cost on GSM-Symbolic, 19% lower cost on augmented GAIA, and $9.5\times$ lower agent latency on $τ^2$-telecom.

cs.AI↗

arXiv2026

PADMÉ: Preference Alignment Data Synthesis for Meta-Evaluation of LM Agent Evaluators

Language models are frequently employed to evaluate other language models. An LM evaluator scoring agentic behaviors across multiple criteria is valuable, provided that its decisions align with human judgment. We call the problem of evaluating this alignment Meta-Evaluation. Tackling it directly is difficult: collecting human data is expensive, absolute scoring is hard to align, and using an LM meta-evaluator recurses the question of trustworthiness. We adopt a reformulation of meta-evaluation as a preference judgment problem: rather than comparing human and LM evaluator scores of a trajectory, we ask whether their implied preferences align. Building on this, we introduce PADMÉ, a data synthesis method that generates reliable criterion-based meta-evaluation data for agentic settings. PADMÉ uses only small language models, requires no human involvement during evaluations, and operates under a low computational budget. We build a prototype of PADMÉ and synthesize a dataset of 1,000 samples across four agentic domains and three evaluation criteria. Human validation on a 150-sample subset demonstrates that PADMÉ improves agreement with human judgment from 73% to 85% over a naive baseline. Meta-evaluating 25 common models with our dataset demonstrates the correlations between evaluation performance and scoring granularity, leniency, and model size, among other factors.

cs.CL↗

arXiv2026

Chatbot Engagement Does Not Always Beget Metalearning: Evidence from Three Countries

Chatbots deliver real-time fact-checks, but whether a chatbot correction leaves anything behind once the chatbot is gone - metalearning, distinct from correcting misbeliefs - is untested. We report a preregistered, three-country randomized experiment (USA, India, Singapore; N ~ 2,200) on out-of-context image misinformation, manipulating a correction's channel affordances (synchronicity, bandwidth) across four conditions: Control, Links-only, Static explanation, and a Socratic Chatbot built on a validated out-of-context detector, with an unaided retest one week later. The Chatbot produced the largest immediate discernment gain (d = 0.097, p = .023). All three interventions reduced sharing of false claims (d ~ -0.12, p < .01). One week later, no advantage persisted: the Chatbot arm declined relative to Control, most sharply in India and Singapore, and in India on claims it never discussed. Decay tracked affordance level and did not vary by country. Engagement mechanisms, we argue, do not substitute for slow AI literacy.

cs.HC↗

arXiv2026

Mitigating GenAI-Powered Evidence Pollution for Out-Of-Context Misinformation Detection

While generative artificial intelligence (GenAI) models have achieved significant success, their misuse for generating deceptive content raises growing concerns about online information security. Out-of-context (OOC) multimodal misinformation detection systems typically rely on Web-retrieved evidence to identify images repurposed in false contexts, but they are increasingly challenged by the presence of GenAI-polluted evidence. Existing work mainly focuses on verifying claims that have undergone stylistic rewriting at the claim level and assume a clean evidence corpus. In this work, we remove this assumption and systematically study the impact of GenAI-driven evidence pollution threat on OOC detection. We show that polluted evidence can degrade the performance of state-of-the-art detectors by more than 9 percentage points. We propose two mitigating strategies, cross-modal evidence reranking and cross-modal claim-evidence reasoning, to address the challenge posed by polluted evidence. Extensive experiments on two benchmark datasets demonstrate that our approaches effectively enhance the robustness of existing OOC detectors amidst polluted evidence. The source code and data are publicly available at https://github.com/YanZehong/GenAI-Evidence-Pollution.

cs.MM↗

arXiv2026

Overconfident and Blind to Details: Fixing Prompt Insensitivity with Abductive Preference Learning

Vision and language models frequently ignore semantically critical input edits, defaulting to pretraining priors. For example, models will confidently assert a five-legged dog has four legs; consequently, on the VLMBias benchmark, GPT 5.2 and Claude Sonnet 4.6 achieve only $4.6\%$ and $0\%$ accuracy, respectively. Existing methods address this problem through building up datasets that covers the underrepresented inputs to tune the policy function $π(y \mid x)$, where $x$ and $y$ refer to input prompts and responses, respectively. However, prompting baselines yield gains of under $3\%$ on VLMBias due to the low probability density of rare prompts. To bypass this bottleneck, we propose \emph{abductive preference learning} to optimize the abductive policy $π(x \mid y)$. We prove this amplifies forward policy improvements by a factor of $q(y)/p(x)$, where $p(\cdot)$ and $q(\cdot)$ denote the marginal probabilities of the prompt and response, yielding the largest gains on the rarest prompts. Furthermore, we demonstrate that for translation invariant pairwise preference learning methods, such as DPO, estimating $π(x \mid y)$ reduces to a structural data swap that compares prompts for a fixed response, requiring no architectural changes. Empirically, abductive preference learning delivers large gains on long-tail prompt sensitivity: on VLMBias, A-DPO raises accuracy from $3\%$ to $44\%$ ($14\times$), outperforming GPT-5.2 ($4.6\%$) and all closed-source VLMs except Gemini~3~Flash; on Inverse-IFEval, Multi-DPOP reaches $65$--$84\%$, surpassing GPT-5 ($73.7\%$) at the 9B scale while preserving IFBench, unlike DPO which degrades it by $8$--$12\%$.

cs.CL↗

arXiv2026

OSWorld 2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks

Existing computer-use benchmarks fail to capture the realism, complexity, and long-horizon demands of real-world computer use, limiting their ability to reveal the limitations of frontier agents. We introduce OSWorld 2.0, a benchmark of 108 long-horizon computer-use workflows across everyday and professional tasks, designed to capture complex and challenging real-world phenomena. Each task represents a realistic end-to-end workflow that takes human users a median of about 1.6 hours to complete and requires an average of 318 tool calls with Claude Opus 4.7 using maximum thinking, compared with about 30 in OSWorld 1.0. OSWorld 2.0 targets challenge phenomena that are common in real workflows yet underrepresented in prior benchmarks, spanning interaction-design challenges such as streaming interaction and dynamic environments, as well as agent-pattern challenges such as cross-source reasoning, implicit-state inference, and visual-spatial precision. Tasks are grounded in authentic input artifacts and cross-referenced against realistic stateful user profile data, and include separate safety reports auditing safety-sensitive execution. Under our primary binary-completion metric at 500 steps, Claude Opus 4.8 with maximum thinking and batched tool calls scores best but still completes only 20.6% of tasks at a 54.8% partial score; GPT-5.5 is far more token-efficient yet plateaus near 13%. These results show that current agents are still far from professional-level computer use: rather than stumbling on basic GUI control or coding, they lose track of constraints, miss information that arrives mid-task, guess rather than ask the user, and skip verification, struggling most when a task hinges on hidden state they must recover.

cs.AI↗

arXiv2026

Learning Expert Strategy for Autonomous Robotic Endovascular Intervention via Decoupled Procedural Execution

Endovascular interventions are high-stakes procedures requiring precise device operation within complex and tortuous vascular anatomies. Autonomous endovascular navigation has the potential to standardize procedural quality and reduce the performance variability inherent in manual operation. Although Reinforcement Learning (RL) approaches have demonstrated promise in enabling autonomy in endovascular intervention, they often struggle with explicit constraint satisfaction and safety guarantees. To address these challenges, a learning-based expert strategy is introduced, enhancing procedural consistency in autonomous endovascular intervention by explicitly decoupling high-level strategic decision-making from low-level procedural execution. The proposed framework replicates the expert clinical decision-making process: a strategic RL policy generates global navigation intents, which are subsequently refined through an expert-informed execution module. This module ensures that robot movements strictly adhere to expert operational norms, real-time kinematic limits, and vessel safety constraints. Experimental evaluation across high-fidelity 3D simulations and a real-world robotic platform demonstrates that the proposed framework not only outperforms baseline policies but also effectively replicates expert-level proficiency. The framework achieves a high navigation success rate (> 96%) and a 29.3% reduction in operational steps, which translates to enhanced operative efficiency and minimized device-vessel interaction. Furthermore, a 13% reduction in trajectory variance indicates superior procedural standardization, aligning autonomous behavior with established clinical norms. These results underscore its potential to enhance the predictability, safety, and consistency of robotic endovascular interventions.

cs.RO↗

arXiv2026

Vision-Language Procedural Reasoning for Context-Aware Reward Modeling of Robotic Endovascular Guidewire Navigation

Robotic-assisted endovascular interventions demand accurate, stable, and context-aware guidewire navigation in complex and patient-specific vascular anatomies. Despite recent advances in robotic precision and learning-based control, existing autonomous navigation methods remain limited by their reliance on static reward functions and the lack of explicit procedural reasoning regarding anatomical context and task progression. To address these challenges, this paper proposes a vision-language procedural reasoning (VL-PR) framework for autonomous guidewire navigation. The framework integrates a multimodal large language model (MLLM) as a procedural reasoning module that interprets real-time visual observations to infer high-level navigation contexts. Instead of generating low-level control commands, the inferred procedural insights enable context-aware reward adaptation by dynamically adjusting the importance of reward components across different navigation phases. This approach allows a single policy to resolve competing objectives and handle complex transitions while preserving a consistent global task goal. Experiments on a physical robotic platform across diverse vascular scenarios demonstrate enhanced task reliability and streamlined navigational efficiency, highlighting the advantages over static-reward methods and offering a scalable solution for complex and multi-task robotic endovascular procedures.

cs.RO↗

arXiv2026

Learning Efficient Guardrails for Compliance

Autonomous web agents are increasingly deployed for long-horizon tasks, yet their ability to adhere to real-world policies remains critically underexplored compared to standard safety objectives. To address this gap, we introduce PolicyGuardBench, a benchmark of 60k policy-trajectory pairs designed to evaluate compliance through both full-trajectory and novel prefix-based violation detection tasks. Using this dataset, we train PolicyGuard, a lightweight guardrail model that achieves strong detection accuracy while maintaining high inference efficiency. Notably, our model demonstrates robust generalization capabilities, preserving high performance even on unseen domains. These contributions establish a comprehensive framework for studying policy compliance, showing that accurate and generalizable guardrails are feasible at small scales.

cs.AI↗

arXiv2026

WARC-Bench: Web Archive Based Benchmark for GUI Subtask Executions

Training web agents to navigate complex, real-world websites requires them to master $\textit{subtasks}$ - short-horizon interactions on multiple UI components (e.g., choosing the correct date in a date picker, or scrolling in a container to extract information). We introduce WARC-Bench (Web Archive Benchmark), a novel web navigation benchmark featuring 438 tasks designed to evaluate multimodal AI agents on subtasks. WARC-Bench enables sandboxed interactions with dynamic and realistic webpages using Web ARChive files. We show that WARC-Bench is challenging for leading computer-use models, with the highest observed success rate being 64.8%. To improve open source models on subtask, we explore two common training techniques: supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). Experiments show that SFT models obtain a 48.8% success rate on the benchmark. Training with RLVR over SFT checkpoints, even in data-scarce settings, improves the score to 52.8% on WARC-Bench, outperforming many frontier models. Our analysis concludes that mastering these subtasks is essential for robust web planning and navigation, and is a capability not extensively evaluated by existing benchmarks.

cs.LG↗

arXiv2026

When Vision Speaks for Sound

Despite rapid progress in video-capable MLLMs, we find that their apparent audio understanding in videos is often vision-driven: models rely on visual cues to infer or hallucinate acoustic information, rather than verifying the audio stream. This issue appears across both state-of-the-art open-source omni models and leading closed-source models from providers such as Google and OpenAI. We characterize this failure mode as an audio-visual Clever Hans effect, in which models appear (falsely) audio-grounded, but actually exploit visual-acoustic correlations without verifying whether the audio and visual streams are truly aligned. To systematically study this behavior, we introduce Thud, an intervention-driven probing framework based on three counterfactual audio edits: Shift, which tests temporal synchronization; Mute, which tests sound existence; and Swap, which tests audio-visual consistency. Beyond diagnosis, we further study a two-stage alignment recipe: intervention-derived preference pairs teach audio verification, while event-level general video preferences regularize the model against over-specialization. Our best 10K-sample recipe improves average performance across the three intervention dimensions by 28 percentage points, while slightly improving performance on general video and audio-visual QA benchmarks.

cs.CV↗

arXiv2026

AutoRubric: Rubric-Based Generative Rewards for Faithful Multimodal Reasoning

Multimodal large language models (MLLMs) have rapidly advanced from perception tasks to complex multi-step reasoning, yet reinforcement learning with verifiable rewards (RLVR) often leads to spurious reasoning since only the final-answer correctness is rewarded. To address this limitation, we propose AutoRubric, a framework that integrates RLVR with process-level supervision through automatically collected rubric-based generative rewards. Our key innovation lies in a scalable self-aggregation method that distills consistent reasoning checkpoints from successful trajectories, enabling problem-specific rubric construction without human annotation or stronger teacher models. By jointly leveraging rubric-based and outcome rewards, AutoRubric achieves state-of-the-art performance on six multimodal reasoning benchmarks and substantially improves reasoning faithfulness in dedicated evaluations.

cs.CL↗

arXiv2026

Directional Dark Field for Nanoscale Full-Field Transmission X-Ray Microscopy

Dark-field X-ray imaging visualizes structural inhomogeneities through small-angle scattering, but existing directional methods are confined to the micrometer scale. While recent advances have extended dark-field capabilities to nanoscale transmission X-ray microscopy, directional scattering retrieval - critical for characterizing anisotropic nanostructures - has remained inaccessible for imaging resolutions in the sub-micrometer scale. Here, we demonstrate the first directional dark-field setup for nanoimaging, achieving orientation mapping of scattering features below the spatial resolution limit. Our method is experimentally simple to implement with existing transmission X-ray microscopy setups. We validate its performance by successfully resolving sub-resolution test structure orientations, cross-correlating orientational changes within hierarchical nanoporous materials, and mapping the directional arrangement of hydroxyapatite nanocrystals 30 - 70 nm within human tooth enamel. By utilizing shadow regions in the optical configuration, we further extend the detectable scattering vector range, demonstrating a pathway toward size-selective dark-field imaging. This advancement enables the quantitative structural characterization of anisotropic nanomaterials, which are critical to biomineralization, advanced materials, and nanotechnology applications.

physics.optics↗

arXiv2026

PolySkill: Learning Generalizable Skills Through Polymorphic Abstraction

Large language models (LLMs) are moving beyond static uses and are now powering agents that learn continually during their interaction with external environments. For example, agents can learn reusable skills while navigating web pages or toggling new tools. However, existing methods for skill learning often create skills that are over-specialized to a single website and fail to generalize. We introduce PolySkill, a new framework that enables agents to learn generalizable and compositional skills. The core idea, inspired by polymorphism in software engineering, is to decouple a skill's abstract goal (what it accomplishes) and its concrete implementation (how it is executed). Experiments show that our method (1) improves skill reuse by 1.7x on seen websites and (2) boosts success rates by up to 9.4% on Mind2Web and 13.9% on unseen websites, while reducing steps by over 20%. (3) In self-exploration settings without specified tasks, our framework improves the quality of proposed tasks and enables agents to learn generalizable skills that work across different sites. By enabling the agent to identify and refine its own goals, the PolySkill enhances the agent's ability to learn a better curriculum, leading to the acquisition of more generalizable skills compared to baseline methods. This work provides a practical path toward building agents capable of continual learning in adaptive environments. Our findings show that separating a skill's goal from its execution is a crucial step toward developing autonomous agents that can learn and generalize across the open web continuously. Our code can be found in https://github.com/simonucl/PolySkill.

cs.CL↗

arXiv2026

Vision-Based Reasoning with Topology-Encoded Graphs for Anatomical Path Disambiguation in Robot-Assisted Endovascular Navigation

Robotic-assisted percutaneous coronary intervention (PCI) is constrained by the inherent limitations of 2D Digital Subtraction Angiography (DSA). Unlike physicians, who can directly manipulate guidewires and integrate tactile feedback with their prior anatomical knowledge, teleoperated robotic systems must rely solely on 2D projections. This mode of operation, simultaneously lacking spatial context and tactile sensation, may give rise to projection-induced ambiguities at vascular bifurcations. To address this challenge, we propose a two-stage framework (SCAR-UNet-GAT) for real-time robotic path planning. In the first stage, SCAR-UNet, a spatial-coordinate-attention-regularized U-Net, is employed for accurate coronary vessel segmentation. The integration of multi-level attention mechanisms enhances the delineation of thin, tortuous vessels and improves robustness against imaging noise. From the resulting binary masks, vessel centerlines and bifurcation points are extracted, and geometric descriptors (e.g., branch diameter, intersection angles) are fused with local DSA patches to construct node features. In the second stage, a Graph Attention Network (GAT) reasons over the vessel graph to identify anatomically consistent and clinically feasible trajectories, effectively distinguishing true bifurcations from projection-induced false crossings. On a clinical DSA dataset, SCAR-UNet achieved a Dice coefficient of 93.1%. For path disambiguation, the proposed GAT-based method attained a success rate of 95.0% and a target-arrival success rate of 90.0%, substantially outperforming conventional shortest-path planning (60.0% and 55.0%) and heuristic-based planning (75.0% and 70.0%). Validation on a robotic platform further confirmed the practical feasibility and robustness of the proposed framework.

cs.RO↗