As a representative form of embodied robotic interaction, aerial grasping requires close coordination between the aerial platform and manipulator. Most existing systems perform aerial grasping with static or slowly moving targets under fixed scenarios. Dynamic and precise grasping across different conditions introduces additional challenges, including target motion, localization drift, dynamic coupling, and external disturbances. In this work, we present CAPTURER, a dexterous aerial manipulator for centimeter-level grasping across diverse environments. To achieve the challenging goal, we first develop an onboard perception system for state estimation and target detection. Drift correction and global pose transformation are introduced to improve target localization during flight. A cooperative motion planner is proposed to coordinate the mobility of the aerial platform with the dexterity of the manipulator according to the relative target state. To maintain grasping accuracy under disturbances, we design a disturbance-rejection controller to compensate for dynamic coupling, model uncertainty, and external disturbances. The capability of CAPTURER is validated through indoor and outdoor experiments with both static and moving targets. The results demonstrate that CAPTURER achieves centimeter-level grasping accuracy for a marker pen measuring only 1.5 cm in diameter at flight speeds exceeding 1.5 m/s.
Memory systems are essential for personalized Large Language Models (LLMs). However, retrieval within these systems primarily relies on semantic similarity, potentially missing logically critical memories with limited semantic overlap. We refer to this failure as semantic deflection, which remains insufficiently evaluated by existing benchmarks. To address this gap, we construct IMLogic, a benchmark for implicit logical memory retrieval in long dialogues. Each question is designed around the divergence between semantic and logical relevance, with diagnostic response options that probe semantic deflection, hallucination, and lack of personalization. To help systems utilize associations that are difficult to discover through semantic matching, we introduce root memory, a structured representation that distills reusable, response-guiding logic from user histories through LLM reasoning and grounds it in supporting evidence. We further develop RootMem, a plug-and-play framework that constructs and routes root memories to complement semantic retrieval. Extensive experiments reveal persistent challenges across retrieval methods and memory systems, including substantial preference-driven errors even under full-memory access. RootMem consistently improves performance, demonstrating the potential of structured personalized logic while leaving considerable challenges unresolved. Our benchmark and code are available at https://github.com/ThisIsCosine/IMLogic.
Evaluating human-AI co-design requires testing whether coding agents can translate real users' multimodal requests into working application revisions. We introduce Co-Design Bench, a reference-free benchmark of 273 tasks curated from 37,463 deduplicated queries across 26,314 real user coding sessions. Its two tracks, Multimodal Editing (ME) and Repository Editing (RE), cover localized edits with optional image, audio, video, or document inputs and coordinated repository revisions. Tasks span three capability dimensions: Grounded Understanding, UI Transformation, and Functional Implementation. The tasks pair normalized requests and runnable pre-edit applications with 2,331 hidden expert-authored acceptance criteria. Our judge-then-refute protocol grades criteria using source and browser evidence, challenges positive verdicts with a separate LLM verifier, and audits visual and interaction claims without gold code or target screenshots. Under a shared agentic harness, the strongest of ten model endpoints passes an average of 69.9% of criteria per task but fully satisfies only 26.4% of tasks. All ten endpoints achieve lower criterion completion on repository edits spanning three or four change types than on those spanning one or two. Co-Design Bench provides a testbed for studying the gap between partial requirement satisfaction and complete task success under agentic execution for real-user application revisions.
Kairui Hu Nanyang Technological University, Penghao Wu Nanyang Technological University, Fanyi Pu, Wang Xiao, Xiang Yue Carnegie Mellon University, Bo Li, Yuanhan Zhang, Ziwei Liu Nanyang Technological University
Humans acquire knowledge through three cognitive stages: perceiving information, comprehending knowledge, and adapting knowledge to solve novel problems. Videos serve as an effective medium for knowledge acquisition, facilitating a natural progression through these learning stages. However, existing video benchmarks fail to evaluate the knowledge acquisition capabilities of Large Multimodal Models (LMMs). To address this gap, we introduce Video-MMMU, a multi-modal, multi-discipline, multi-track benchmark that evaluates LMMs’ ability to acquire knowledge from college-level, educational videos. Video-MMMU features a collection of 300 videos and 900 human-annotated questions across six disciplines, evaluating knowledge acquisition through stage-aligned question-answer pairs: Perception, Comprehension, and Adaptation. Beyond measuring final accuracy, Video-MMMU proposes the performance gain metric that quantifies an LMM’s learning gain from video, shifting the focus of evaluation from absolute performance to learning efficiency. Our evaluation reveals a substantial gap between human learners and current LMMs, highlighting the need to improve models’ ability to learn and adapt knowledge from video content.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Yuetai Li, Zhangchen Xu, Fengqing Jiang Apple and University of Washington, Bhaskar Ramasubramanian Western Washington University, Luyao Niu University of Washington, Bill Yuchen Lin xAI and University of Washington, Xiang Yue Carnegie Mellon University, Radha Poovendran University of Washington, Seattle
Fine-tuning large language models (LLMs) is intended to improve their reasoning capabilities, yet we uncover a counterintuitive effect: models often forget how to solve problems they previously answered correctly during training. We term this phenomenon Temporal Forgetting and show that it is widespread across model sizes, fine-tuning methods (both Reinforcement Learning and Supervised Fine-Tuning), and multiple reasoning benchmarks. Our analysis reveals on average more than 20% of final errors were once solved correctly at an earlier checkpoint. Inspired by the phenomenon of Temporal Forgetting, we proposed Temporal Sampling, a simple decoding strategy that draws outputs from multiple checkpoints along the training trajectory. This approach recovers forgotten solutions and leads to significant improvements in reasoning performance than final-ckpt-sampling only, gains from 4 to 19 points in Pass@k and consistent gains for majority-voting and Best-of-N across several benchmarks. Temporal sampling also outperforms strong baselines such as model merging. By leveraging the temporal diversity inherent in training, Temporal Sampling offers a practical, compute-efficient way to surface hidden reasoning ability and rethink how we evaluate LLMs.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表