The paper presents a generative framework that estimates category-level 6D pose and 3D size of objects from a single RGB image, using score-based diffusion models to produce a multi-hypothesis pose distribution. It replaces costly likelihood pruning with a Mean Shift approach to isolate the mode as the final pose estimate, achieving state-of-the-art results on the REAL275 benchmark. The method also decouples detection from pose estimation, enabling robust zero-shot generalisation on the Wild6D dataset and extending naturally to video sequences by propagating the pose distribution over time.
By Adam Bethell, Ravi Garg, Ian Reid
RideWay is a new benchmark that evaluates ride‑hailing language agents not just on task completion but on interaction efficiency. It introduces the Efficiency Utility metric, which penalizes agents for excessive tool calls and user‑facing turns relative to a task‑specific reference effort, with human preferences used to calibrate the penalties. Across 58 tasks and 24 models, the metric shows that extra dialogue is penalized more heavily than extra tool use, and it achieves high accuracy in distinguishing trajectories that differ in turns but struggles when differences are only in tool calls.
By Qingnuan Han, Boli Fang, Mingzhi Hou, Claire Liu
GraphEcho is a benchmark that examines how large language model agents navigate graph paths and handle evidence redundancy. It tests whether agents treat repeated encounters as additional corroboration by varying path counts and evidential origins while keeping evidence content constant. The study finds that redundant paths increase repeated walks, and that provenance-aware post‑training can reduce revisits but may limit source diversity, revealing a gap between efficient exploration and effective evidence use.
By Sikun Wang, Yixi Zhou, Lei Fan, Fan Zhang
The paper introduces a benchmark for recommending SNOMED CT concepts from masked clinical contexts, using data from the SNOMED CT Entity Linking Challenge v1.2.1 derived from MIMIC-IV-Note. It evaluates several methods—including a popularity baseline, sparse TF‑IDF prototypes, dense embeddings, and retrieval‑augmented hybrids—finding that sparse TF‑IDF achieves the best performance with Recall@1 of 14.81% and Recall@10 of 33.43%. The study highlights that concept frequency and lexical context strongly influence recommendation quality, with many test pairs involving concepts unseen during training.
By Ali Noori
The paper proposes an epidemic model to explain how a multi‑agent system can shift from a single accidental deviation to a collective loss of control. It identifies accidental mutation, contagion through communication, and recovery as key mechanisms, and demonstrates that unsafe trajectories can spread rapidly among agents, leading to high harm rates in injected scenarios. The study also highlights the importance of auditing communication paths and strengthening both prevention and recovery measures to mitigate such risks.
By Xiangfan Wu, Zonghao Ying, Huiyu Wu, Xing Zheng, Huangsheng Cheng, Xiaorong Shi, Jing Guo
RiskWorld is a risk‑aware world modeling framework that forecasts shared occupancy and selectively replaces planned trajectories in automated driving. It fuses spatial risk fields, temporal actor context, and visual bird’s‑eye‑view features, using flow‑guided evolution to transport occupancy and signed residuals to correct it. In open‑loop planning on nuScenes, RiskWorld achieves the lowest collision rate over a 3‑second horizon and the second‑best average L2 error, running at 11.5 FPS on a single NVIDIA RTX 4090.
By Rongxiang Zeng, Linsen Cai, Jiafu Zhang, Yijie Zhong, Yide Tao, Shuai Wang, Nan Zheng, Hai L. Vu, Alvaro Garcia Hernandez, Yongqi Dong
The paper investigates why large reasoning models (LRMs) lose safety alignment when faced with harmful queries. By analyzing token-level refusal dynamics, the authors identify a vulnerability called Onset Refusal Collapse (ORC), where the refusal signal drops sharply at the first generated token, leading to unsafe responses. They introduce SafeToken, a lightweight inference-time intervention that injects a learned safety anchor at reasoning onset, which mitigates ORC, improves safety on harmful-query benchmarks, and largely preserves reasoning utility.
By Yizheng Yang, Haining Yu, Yuechen Wang, Yikai Hou, Xing Fu, Jinbo Yang, Tianqing Zhu
Linear probes can decode safety‑relevant concepts such as truthfulness from language‑model activations, but probe accuracy may reflect only decodability, not causal influence on model behavior. The authors show that probe weight geometry alone cannot identify the features the model actually uses, because geometrically aligned features need not be causally relevant. They introduce a sparse‑autoencoder (SAE) decomposition that ranks features by probe alignment and gradient sensitivity, and demonstrate that ablating shared, probe‑only, and random feature sets reveals a sharp dissociation: shared features drive model output changes far more than probe‑only or random features, confirming that causal relevance requires intervention beyond weight geometry.
By Devesh Tiwari, Camille Davis, Shivank Sinha, Talia Weaver, Aditya Shah, Maheep Chaudhary
ContrAgent is a contract‑based framework that provides symbolic temporal supervision for large language model agents. It records an agent’s tool‑call sequence as a trace of checkable predicates and formalizes desired behaviors with assume‑guarantee contracts expressed in linear temporal logic over finite traces (LTLf). Each contract is compiled into a deterministic finite automaton that both gates actions online and evaluates recorded traces offline, enabling deterministic, reproducible verdicts and significantly lower per‑call latency compared to existing LLM‑judge and rule‑based guardrail baselines.
By Yifeng Xiao, Pierluigi Nuzzo
Re2A is a new framework for situated conversational recommendation that models user interactions within shared physical environments. It introduces rubric-based preference reasoning to explicitly capture user preferences from dialogue history and scene context, and a preference-conditioned optimization to align generated responses with both user satisfaction and situational consistency. Experiments on two SCR datasets show that Re2A outperforms existing methods, providing more precise and context-aware recommendations.
By Dongding Lin, Jian Wang, Xiaoyan Zhao, Wenjie Li
BENCHCOMPASS is a new payment‑domain benchmark that transforms typed evidence packs into scenario‑grounded tasks, applies LLM‑based quality checks, generates attack variants, and reserves final item admission for domain experts. It includes an expert‑reviewed Pro benchmark covering payment knowledge, context‑grounded scenario reasoning, and attacked open robustness, plus a lower‑assurance Normal pool. Across 16 model variants, BENCHCOMPASS reveals distinct failure modes—missing payment knowledge, incomplete reasoning, and failure to reject invalid workflows—while the best model scores 89.6% on open context‑grounded reasoning and 81.7% under attacked inputs.
"whyItMatters":"The benchmark provides a structured way to isolate and evaluate specific weaknesses in LLMs for payment operations, a critical financial infrastructure where rules change rapidly and decisions depend on complex contextual factors."
By Sijie Dong, Wei Ren, Xuanwei Hu, Jiawei Luo, Zifan Wang, Xiaoyun Feng, Hui Cai, Lyuxin Xue, Peng Lu, Jianshe Li, Xin Zhang, Wei Wu
The paper introduces "cunning questions"—non‑safety prompts that contain misleading premises or subtle inconsistencies—to train large language models (LLMs) to scrutinize underlying intent and assumptions. Experiments show that incorporating these questions improves robustness against out‑of‑distribution jailbreak attacks and enhances subsequent safety fine‑tuning, achieving a new state‑of‑the‑art reduction in mean ASR from 17.40% to 15.05% across nine backbone–benchmark combinations. The authors argue that this training fosters vigilance, enabling models to prioritize safety judgments before engaging in harmful planning.
By Youjia Wang, Lin Xu, Yang Sun, Yuxiao Lu, Chengfang Fang, Jie Shi
The paper evaluates how well current Large Language Models can translate natural language goals, written by video game testers, into well‑formed PDDL targets for classical planning. Using a carefully designed prompt template, six state‑of‑the‑art LLMs were tested on correctness, speed, and error tendencies with real‑world benchmarks. All models achieved high correctness (>92%), with Gemini 2.5 Flash reaching 96% accuracy and the fewest false positives, while GPT‑4.1 was the fastest, yet differences in performance and occasional failures due to ambiguity and domain limits remain.
By Tomas Balyo, Lukas Chrpa, G. Michael Youngblood
CERA-MoA introduces a co-evolving router and independently learning agents for Mixture-of-Agents systems, allowing routing strategies to adapt to evolving agent capabilities. The framework uses a predictive familiarity estimator based on mid-layer hidden states to evaluate semantic competence, and a cumulative-threshold adaptive routing mechanism to activate a minimal agent subset. Experiments across multiple domains show that CERA-MoA outperforms static-agent routing and fixed-workflow fine-tuning baselines.
By Jiaxuan Jiang, Liyuan He, Zhixuan Fang
REVERSAL-BENCH is a benchmark that introduces a continuous reversibility parameter ρ∈[0,1] and a reset oracle to evaluate how well reinforcement learning agents can recover from irreversible states across eight manipulation tasks in five physics engines. Experiments show a sharp reversibility cliff: reset‑free agents become trapped in irrecoverable states as ρ increases, while episodic agents continue learning steadily. The benchmark also provides a large multi‑simulator dataset and demonstrates that safety shields can predict recoverability but only succeed when the agent can avoid the trap.
By Riyaaz Shaik, Chandru Venkataraman
The paper revisits Thompson’s classic compiler back‑door attack in the context of self‑modifying AI coding agents. By poisoning the benchmarks used for self‑evaluation, the authors demonstrate that agents such as the Darwin Gödel Machine, Self‑Improving Coding Agent, and Hyperagents can be coaxed into generating vulnerable code, even on clean, held‑out tasks. Experiments show that the contamination can persist after subsequent clean training, highlighting the need for more robust agent designs.
By Franziska Roesner, Tadayoshi Kohno
The paper introduces GYROval, a benchmark designed to assess cultural value orientation in large language models along the Inglehart‑Welzel axes across multiple domains and roles. It uses binary contrastive scenarios where both options are valid, and scores models by the proportion of responses aligning with each pole. The benchmark was tested on twenty models, with eleven also evaluated in Russian and at a different sampling temperature, and its stability was analyzed using tie‑corrected Kendall’s W against a permutation null.
By Alexander Didenko, Anna Shabanova, Vladislav Zapylikhin, Alexander Antipov, Ruslana Raemgulova
The paper introduces Mask 2D-3D, an Adaptive Dual-Masked Autoencoder Network designed for image-to-point cloud registration. It proposes an Intermodal Dual-MAE Framework (ID-MAE) with a Similarity-based RL Masking Strategy (SRLM) that adaptively masks informative positions using cross-modal similarity and reinforcement learning. Experiments on RGB-D Scenes v2 and 7-Scenes benchmarks demonstrate state-of-the-art performance in this registration task.
By Zhixin Cheng, Jiacheng Deng, Xiaotian Yin, Baoqun Yin, Richang Hong, Tianzhu Zhang
The paper reports the first systematic audit of open‑weight large language models (LLMs) in hiring contexts, examining how job‑posting language influences recruiter and job‑seeker simulations across six models. It finds that agentic language lowers recruiter scores for female candidates while communal language mitigates this effect, and that coded‑exclusion language sharply reduces recruiter scores for non‑White candidates and discourages non‑White personas from applying. The study also identifies the explicit demographic label as the main causal factor and proposes a concrete pre‑deployment audit protocol aligned with EU and U.S. regulatory requirements.
By Kosuke Kitahara, Nobuhiro Yamaguchi
The paper derives a closed‑form expression for the reduced quantum channel of a single‑qubit Interactive Quantum Classifier (IQC), revealing that Hamiltonian parameters directly control the constant, sine, and cosine components of the classifier output. This Fourier interpretation motivates a generalized family of Hamiltonian encodings, including matrix‑parameterized environmental Hamiltonians that produce non‑separable Fourier structures. Numerical experiments on synthetic and real‑world datasets demonstrate that these generalized encodings improve classification performance on several nonlinear benchmarks, with a simpler four‑parameter extension achieving comparable results with fewer trainable parameters.
By F\'abio Novaes, Fernando M. de Paula Neto, Jo\~ao V. M. Cardoso