arXiv subjects

Jiajie Zhang

Publications and source records attributed to Jiajie Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Memory That Changes Action Is Not Memory That Guides It: Counterfactual Auditing of History-Conditioned Robot Policies

A robot returning a block to its origin tray may encounter two task-consistent pasts that reconverge to the same current input but warrant different actions. Yet memory-policy evaluations often rely on task success or action change under memory perturbation, neither of which establishes that memory guides the decision. We propose the \textbf{Counterfactual Memory Audit (CMA)}, an evaluation protocol that crosses two histories at a verified-identical present, queries a frozen policy under common randomness, and evaluates each saved action under both pasts. This separates memory sensitivity, warranted choice, matched-world physical value, and per-pair reliability. On Mem-0, every audited Put Back pair changes action, but only $20/64$ pairs are fully reliable; at a later Swap decision, all paired actions change while both memories select the same branch. Native interventions further show closed-loop influence: replacing the history bank redirects behavior toward the replaced content, while restoring a 4096-byte protected anchor recovers $38.9$ points of Swap success lost to injected bank faults. On a dual-arm physical platform, memory changes saved actions, yet five of nine completed Put Back manipulations reach the wrong target. These results show that a robot can remember and react without reliably using memory to choose the behavior its past warrants. CMA provides a decision-level audit for distinguishing these cases.

cs.RO↗

arXiv2026

Learning Choice Model Trees for Feature-Based Multi-Product Pricing: Exact Optimization and Field Evidence

Feature-based multi-product pricing uses customer characteristics to identify demand heterogeneity and tailor prices across products. Choice model trees segment customers through interpretable feature rules and fit a demand model within each leaf. Existing methods typically construct these trees greedily, selecting one myopic split at a time. We develop optimal choice model trees with multinomial logit leaves (OCMT-MNL), jointly optimizing the tree and leaf models within a prescribed depth. Our exact dynamic program derives closed-form Fenchel lower bounds during constrained Newton iterations and propagates them across nested and disjoint customer subsets, avoiding new fits and resuming unfinished fits without repeating completed work. In synthetic experiments, it reduces exact leaf fits by 99.98% and leaf evaluations by 86.13%, achieving up to 7.15-fold speedups over unpruned dynamic programming. One-dimensional lookup tables translate offline estimation into real-time pricing, with a revenue-loss bound quadratic in grid spacing under the fitted model. Compared with greedy trees, OCMT-MNL achieves lower revenue loss with fewer leaves on synthetic data and better predictive fit on real data. In a 23-week randomized experiment on ancillary seat pricing across 48 airline markets and 190,220 passengers, OCMT-MNL increases seat revenue per passenger by a statistically significant 11.3% over static pricing.

math.OC↗

arXiv2026

PathScale-R1: Cross-scale Reasoning for Pathological Image Analysis

Pathological diagnosis is inherently multi-scale, requiring the integration of global tissue architecture at low magnification with cellular morphology at higher magnification. However, existing pathology benchmarks and vision-language models (VLMs) are still largely developed under single-scale settings, limiting their ability to learn clinically meaningful multi-magnification reasoning. Moreover, naively constructed visual question answering (VQA) tasks may be susceptible to text-only or superficial visual shortcuts, leading to unreliable assessments of visual understanding. To address these limitations, we introduce a benchmark and training framework for shortcut-resistant cross-scale pathology reasoning. We design an Adversarial Text-only Screening strategy for semantic reasoning questions and a Structure-controlled Distractor Sampling strategy for visual grounding questions, encouraging models to rely on cross-scale visual evidence. Based on this pipeline, we construct PathScale-VQA, a high-quality cross-scale pathology VQA benchmark with 10,373 multiple-choice questions grounded in 1,368 diagnostic paths across multiple magnification levels. Building on the semantic reasoning set, PathScale-R1 is optimized through Difficulty-driven Reasoning Distillation supervised fine-tuning followed by reinforcement learning with a Scale-aware Reasoning Structure reward, which encourages the use of evidence across magnifications. Extensive experiments demonstrate state-of-the-art performance of PathScale-R1 on cross-scale reasoning tasks and effective transfer to conventional single-scale pathology VQA. Our code is available at https://github.com/iMVR-PL/PathScale-R1.

cs.CV↗

arXiv2026

LLM-Driven Approach to Modeling Tool Interoperability in Automotive Domain

Interoperability between heterogeneous modeling tools remains a significant challenge in Model-Driven Engineering (MDE), particularly in the automotive domain where multiple modeling languages, as well as defacto standard proprietary and open-source tools coexist. This paper presents an LLM-driven approach for automated model interoperability by considering two relevant aspects: 1) mapping model instances to a target metamodel 2) merging of metamodels. The proposed methodology is demonstrated through transformations involving Ecore and SysML v2 based metamodels and incorporates structural validation of generated model instances against user-defined target models. Automotive case studies illustrate the feasibility of the approach and show that large language models can significantly reduce manual transformation effort while generating structurally valid target models for cross-tool interoperability.

cs.SE↗

arXiv2026

Ultra-High-Energy Cosmic Ray Boosted Relic Neutrinos

Ultra-high-energy cosmic rays (UHECRs) can boost relic neutrinos to high energies through Standard Model (SM) neutral-current interactions, providing an indirect probe of the cosmic neutrino background (C$ν$B). In this work, we perform a systematic study of the diffuse UHECR-boosted C$ν$B flux including elastic neutrino-nucleon scattering (ES), coherent elastic neutrino-nucleus scattering (COH), incoherent neutrino-nucleus scattering (INCOH), baryon-resonance production (RES), and deep inelastic scattering (DIS). For the UHECR flux, we use mixed-composition spectra obtained from the UHECR propagation code PriNCe and from the H3a and H4a implementations of the Hillas model, together with SFR, QSO and GRB source evolution models. We find a clear hierarchy of scattering channels in boosted neutrino energy. The coherent scattering dominates at low-energy neutrino flux for heavy nuclear component, while ES and INCOH become important once individual nucleons are resolved. The RES channel gives a non-negligible contribution in the high-energy region, and DIS appears only at the highest energies and is most visible for the H4a models. Using current IceCube and Pierre Auger Observatory data, we derive upper limits on the C$ν$B overdensity. Our results show that reliable predictions of the UHECR-boosted C$ν$B signal require a combined treatment of the relevant SM scattering channels, UHECR composition, source evolution and the neutrino mass spectrum.

hep-ph↗

arXiv2026

Boundary-Guided Policy Optimization for Memory-efficient RL of Diffusion Large Language Models

A key challenge in applying reinforcement learning (RL) to diffusion large language models (dLLMs) is the intractability of their likelihood functions, which are essential for the RL objective, necessitating corresponding approximation during training. While existing methods approximate the log-likelihoods by their evidence lower bounds (ELBOs) via customized Monte Carlo (MC) sampling, they incur significant memory overhead due to the need to retain all MC samples for the gradient computation of non-linear terms in the RL objective, and thus restrict feasible sample sizes, leading to imprecise likelihood approximations and distorted RL objective. To address this, we propose \emph{Boundary-Guided Policy Optimization} (BGPO), a memory-efficient RL algorithm that maximizes a specially constructed lower bound of the ELBO-based objective. This lower bound is carefully designed to satisfy two key properties: (1) Linearity: it is a linear sum where each term depends only on a single MC sample, thereby enabling gradient accumulation across samples and ensuring constant memory usage; (2) Equivalence: Both the value and gradient of this lower bound are equal to those of the ELBO-based objective in on-policy training, making it also an effective approximation for the original RL objective. These properties allow BGPO to adopt a large MC sample size, improving likelihood approximations and RL objective estimation, which in turn leads to enhanced performance. Experiments show that BGPO significantly outperforms previous RL algorithms for dLLMs in math problem solving, code generation, and planning tasks. Our codes and models are available at \href{https://github.com/THU-KEG/BGPO}{https://github.com/THU-KEG/BGPO}.

cs.LG↗

arXiv2026

LongTraceRL: Learning Long-Context Reasoning from Search Agent Trajectories with Rubric Rewards

Long-context reasoning remains a central challenge for large language models, which often fail to locate and integrate key information in extensive distracting content. Reinforcement learning with verifiable rewards (RLVR) has shown promise for this task, yet existing methods are limited by low-confusability distractors and sparse, outcome-only reward signals that cannot supervise intermediate reasoning steps. To address these issues, we introduce \textsc{LongTraceRL}. For data construction, we generate multi-hop questions via knowledge graph random walks and leverage search agent trajectories to build \emph{tiered distractors}: documents the agent read but did not cite (high confusability) and documents that appeared in search results but were never opened (low confusability), producing training contexts that are far more challenging than those built by random sampling or one-shot search. For reward design, we propose a \emph{rubric reward} that uses the gold entities along each reasoning chain as fine-grained, entity-level process supervision. This rubric reward is applied only to responses with correct final answers (positive-only strategy), distinguishing the reasoning quality among correct responses and preventing reward hacking. Experiments on three reasoning LLMs (4B--30B) across five long-context benchmarks demonstrate that \textsc{LongTraceRL} consistently outperforms strong baselines and encourages comprehensive, evidence-grounded reasoning. Codes, datasets and models are available at \href{https://github.com/THU-KEG/LongTraceRL}{https://github.com/THU-KEG/LongTraceRL}.

cs.CL↗

arXiv2026

SGTA: Scene-Graph Based Multi-Modal Traffic Agent for Video Understanding

We present Scene-Graph Based Multi-Modal Traffic Agent (SGTA), a modular framework for traffic video understanding that combines structured scene graphs with multi-modal reasoning. It constructs a traffic scene graph from roadside videos using detection, tracking, and lane extraction, followed by tool-based reasoning over both symbolic graph queries and visual inputs. SGTA adopts ReAct to process interleaved reasoning traces from large language models with tool invocations, enabling interpretable decision-making for complex video questions. Experiments on selected TUMTraffic VideoQA dataset sample demonstrate that SGTA achieves competitive accuracy across multiple question types while providing transparent reasoning steps. These results highlight the potential of integrating structured scene representations with multi-modal agents for traffic video understanding.

cs.CV↗

arXiv2026

Generation of Indoor Open Street Maps for Robot Navigation from CAD Files

The deployment of autonomous mobile robots is predicated on the availability of environmental maps, yet conventional generation via SLAM (Simultaneous Localization and Mapping) suffers from significant limitations in time, labor, and robustness, particularly in dynamic, large-scale indoor environments where map obsolescence can lead to critical localization failures. To address these challenges, this paper presents a complete and automated system for converting architectural Computer-Aided Design (CAD) files into a hierarchical topometric OpenStreetMap (OSM) representation, tailored for robust life-long robot navigation. Our core methodology involves a multi-stage pipeline that first isolates key structural layers from the raw CAD data and then employs an AreaGraph-based topological segmentation to partition the building layout into a hierarchical graph of navigable spaces. This process yields a comprehensive and semantically rich map, further enhanced by automatically associating textual labels from the CAD source and cohesively merging multiple building floors into a unified, topologically-correct model. By leveraging the permanent structural information inherent in CAD files, our system circumvents the inefficiencies and fragility of SLAM, offering a practical and scalable solution for deploying robots in complex indoor spaces. The software is encapsulated within an intuitive Graphical User Interface (GUI) to facilitate practical use. The code and dataset are available at https://github.com/jiajiezhang7/osmAG-from-cad.

cs.RO↗

arXiv2026

From Observation to Action: Latent Action-based Primitive Segmentation for VLA Pre-training in Industrial Settings

We present a novel unsupervised framework to unlock vast unlabeled human demonstration data from continuous industrial video streams for Vision-Language-Action (VLA) model pre-training. Our method first trains a lightweight motion tokenizer to encode motion dynamics, then employs an unsupervised action segmenter leveraging a novel "Latent Action Energy" metric to discover and segment semantically coherent action primitives. The pipeline outputs both segmented video clips and their corresponding latent action sequences, providing structured data directly suitable for VLA pre-training. Evaluations on public benchmarks and a proprietary electric motor assembly dataset demonstrate effective segmentation of key tasks performed by humans at workstations. Further clustering and quantitative assessment via a Vision-Language Model confirm the semantic coherence of the discovered action primitives. To our knowledge, this is the first fully automated end-to-end system for extracting and organizing VLA pre-training data from unstructured industrial videos, offering a scalable solution for embodied AI integration in manufacturing.

cs.CV↗

arXiv2026

osmAG-Nav: A Hierarchical Semantic Topometric Navigation Stack for Robust Lifelong Indoor Autonomy

The deployment of mobile robots in large-scale, multi-floor environments demands navigation systems that achieve spatial scalability without compromising local kinematic precision. Traditional navigation stacks, reliant on monolithic occupancy grid maps, face severe bottlenecks in storage efficiency, cross-floor reasoning, and long-horizon planning. To address these limitations, this paper presents osmAG-Nav, a complete, open-source ROS2 navigation stack built upon the hierarchical semantic topometric OpenStreetMap Area Graph (osmAG) map standard. The system follows a "System of Systems" architecture that decouples global topological reasoning from local metric execution. A Hierarchical osmAG planner replaces dense grid searches with an LCA-anchored pipeline on a passage-centric graph whose edge costs derive from local raster traversability rather than Euclidean distance, yielding low-millisecond planning on long campus-scale routes. A Rolling Window mechanism rasterizes a fixed-size local metric grid around the robot, keeping the local costmap memory footprint independent of the total mapped area, while a Segmented Execution strategy dispatches intermediate goals to standard ROS2 controllers for smooth handoffs. System robustness is reinforced by a structure-aware LiDAR localization framework that filters dynamic clutter against permanent architectural priors. Extensive experiments on a real-world multi-story indoor-outdoor campus (>11,025 m^2) show that, on the same-floor benchmark subset, osmAG-Nav delivers up to 7816x lower planning latency than a grid-based baseline on long routes while maintaining low path-length overhead and lifelong localization stability. A single-floor long-range robot mission further validates the integrated stack reliability. The full stack is released as modular ROS2 Lifecycle Nodes.

cs.RO↗

arXiv2026

Digital-Twin Losses for Lane-Compliant Trajectory Prediction at Urban Intersections

Accurate and safety-conscious trajectory prediction is a key technology for intelligent transportation systems, especially in V2X-enabled urban environments with complex multi-agent interactions. In this paper, we created a digital twin-driven V2X trajectory prediction pipeline that jointly leverages cooperative perception from vehicles and infrastructure to forecast multi-agent motion at signalized intersections. The proposed model combines a Bi-LSTM-based generator with a structured training objective consisting of a standard mean squared error (MSE) loss and a novel twin loss. The twin loss encodes infrastructure constraints, collision avoidance, diversity across predicted modes, and rule-based priors derived from the digital twin. While the MSE term ensures point-wise accuracy, the twin loss penalizes traffic rule violations, predicted collisions, and mode collapse, guiding the model toward scene-consistent and safety-compliant predictions. We train and evaluate our approach on real-world V2X data sent from the intersection to the vehicle and collected in urban corridors. In addition to standard trajectory metrics (ADE, FDE), we introduce ITS-relevant safety indicators, including infrastructure and rule violation rates. Experimental results demonstrate that the proposed training scheme significantly reduces critical violations while maintaining comparable prediction accuracy and real-time performance, highlighting the potential of digital twin-driven multi-loss learning for V2X-enabled intelligent transportation systems.

cs.RO↗

arXiv2026

GLM-5: from Vibe Coding to Agentic Engineering

We present GLM-5, a next-generation foundation model designed to transition the paradigm of vibe coding to agentic engineering. Building upon the agentic, reasoning, and coding (ARC) capabilities of its predecessor, GLM-5 adopts DSA to significantly reduce training and inference costs while maintaining long-context fidelity. To advance model alignment and autonomy, we implement a new asynchronous reinforcement learning infrastructure that drastically improves post-training efficiency by decoupling generation from training. Furthermore, we propose novel asynchronous agent RL algorithms that further improve RL quality, enabling the model to learn from complex, long-horizon interactions more effectively. Through these innovations, GLM-5 achieves state-of-the-art performance on major open benchmarks. Most critically, GLM-5 demonstrates unprecedented capability in real-world coding tasks, surpassing previous baselines in handling end-to-end software engineering challenges. Code, models, and more information are available at https://github.com/zai-org/GLM-5.

cs.LG↗

arXiv2026

Impact of coherent scattering on relic neutrinos boosted by cosmic rays

Ultra-high-energy cosmic rays (UHECR) scattering off the cosmic relic neutrino background have recently gained renewed interest in the literature. Current data suggest that (UHECR) are predominantly made of heavy nuclei. Similar to the coherent elastic neutrino-nucleus scattering (CE$ν$NS) observed at low-energy neutrino experiments, the cross section of heavy nucleus scattering off relic neutrinos will be coherently enhanced since the energy of relic neutrinos can reach $\sim O(10)$ MeV in the rest frame of the UHECR. We calculate the diffuse flux of relic neutrinos boosted by UHECR after taking into account the contributions from both coherent and incoherent scatterings. Using current data from IceCube and Pierre Auger Observatory, we place constraints on the overdensity of relic neutrinos down to $\sim 10^8$. Since the flux of boosted relic neutrinos peaks at an energy of $\sim 200\, \text{PeV}$, we also entertain the possibility to explain the recently observed KM3NeT event with boosted relic neutrinos from UHECR.

hep-ph↗

arXiv2026

Cross-Level Sensor Fusion with Object Lists via Transformer for 3D Object Detection

In automotive sensor fusion systems, smart sensors and Vehicle-to-Everything (V2X) modules are commonly utilized. Sensor data from these systems are typically available only as processed object lists rather than raw sensor data from traditional sensors. Instead of processing other raw data separately and then fusing them at the object level, we propose an end-to-end cross-level fusion concept with Transformer, which integrates highly abstract object list information with raw camera images for 3D object detection. Object lists are fed into a Transformer as denoising queries and propagated together with learnable queries through the latter feature aggregation process. Additionally, a deformable Gaussian mask, derived from the positional and size dimensional priors from the object lists, is explicitly integrated into the Transformer decoder. This directs attention toward the target area of interest and accelerates model training convergence. Furthermore, as there is no public dataset containing object lists as a standalone modality, we propose an approach to generate pseudo object lists from ground-truth bounding boxes by simulating state noise and false positives and negatives. As the first work to conduct cross-level fusion, our approach shows substantial performance improvements over the vision-based baseline on the nuScenes dataset. It demonstrates its generalization capability over diverse noise levels of simulated object lists and real detectors.

cs.CV↗

arXiv2026

Chaining the Evidence: Robust Reinforcement Learning for Deep Search Agents with Citation-Aware Rubric Rewards

Reinforcement learning (RL) has emerged as a critical technique for enhancing LLM-based deep search agents. However, existing approaches primarily rely on binary outcome rewards, which fail to capture the comprehensiveness and factuality of agents' reasoning process, and often lead to undesirable behaviors such as shortcut exploitation and hallucinations. To address these limitations, we propose \textbf{Citation-aware Rubric Rewards (CaRR)}, a fine-grained reward framework for deep search agents that emphasizes reasoning comprehensiveness, factual grounding, and evidence connectivity. CaRR decomposes complex questions into verifiable single-hop rubrics and requires agents to satisfy these rubrics by explicitly identifying hidden entities, supporting them with correct citations, and constructing complete evidence chains that link to the predicted answer. We further introduce \textbf{Citation-aware Group Relative Policy Optimization (C-GRPO)}, which combines CaRR and outcome rewards for training robust deep search agents. Experiments show that C-GRPO consistently outperforms standard outcome-based RL baselines across multiple deep search benchmarks. Our analysis also validates that C-GRPO effectively discourages shortcut exploitation, promotes comprehensive, evidence-grounded reasoning, and exhibits strong generalization to open-ended deep research tasks. Our code and data are available at https://github.com/THUDM/CaRR.

cs.CL↗

arXiv2025

MMGeoLM: Hard Negative Contrastive Learning for Fine-Grained Geometric Understanding in Large Multimodal Models

Large Multimodal Models (LMMs) typically build on ViTs (e.g., CLIP), yet their training with simple random in-batch negatives limits the ability to capture fine-grained visual differences, particularly in geometric scenarios. To address this challenge, we propose a novel hard negative contrastive learning framework for the vision encoder, which combines image-based contrastive learning using generation-based hard negatives created by perturbing diagram generation code, and text-based contrastive learning using rule-based negatives derived from modified geometric descriptions and retrieval-based negatives selected based on caption similarity. We train a vision encoder (CLIP) using our hard negative training method, namely MMCLIP (Multimodal Math CLIP), and subsequently train an LMM for geometric problem-solving. Experiments show that our trained model, MMGeoLM, significantly outperforms other open-source models on three geometric reasoning benchmarks. Even with a size of 7B, it can rival powerful closed-source models like GPT-4o. We further conduct ablation studies to analyze three key factors: hard negative types, the efficiency of image-based negatives, and training configurations. These analyses yield important insights into optimizing the training pipeline of vision encoder for fine-grained geometric reasoning tasks. https://github.com/THU-KEG/MMGeoLM.

cs.CV↗

arXiv2025

WiFi-based Global Localization in Large-Scale Environments Leveraging Structural Priors from osmAG

Global localization is essential for autonomous robotics, especially in indoor environments where the GPS signal is denied. We propose a novel WiFi-based localization framework that leverages ubiquitous wireless infrastructure and the OpenStreetMap Area Graph (osmAG) for large-scale indoor environments. Our approach integrates signal propagation modeling with osmAG's geometric and topological priors. In the offline phase, an iterative optimization algorithm localizes WiFi Access Points (APs) by modeling wall attenuation, achieving a mean localization error of 3.79 m (35.3\% improvement over trilateration). In the online phase, real-time robot localization uses the augmented osmAG map, yielding a mean error of 3.12 m in fingerprinted areas (8.77\% improvement over KNN fingerprinting) and 3.83 m in non-fingerprinted areas (81.05\% improvement). Comparison with a fingerprint-based method shows that our approach is much more space efficient and achieves superior localization accuracy, especially for positions where no fingerprint data are available. Validated across a complex 11,025 &m^2& multi-floor environment, this framework offers a scalable, cost-effective solution for indoor robotic localization, solving the kidnapped robot problem. The code and dataset are available at https://github.com/XuMa369/osmag-wifi-localization.

cs.RO↗