The paper investigates whether centralized learning improves coordination among embodied agents by comparing independent Q‑learning (IQL) and central Q‑learning (CQL) in a predator‑prey environment. Experiments on 5x5 and 8x8 grids with varying speed and stamina show that CQL does not enhance coordination and can even harm performance; its only advantage is a 42% reduction in capture time on the small grid when the best joint action is explored. The study concludes that tabular comparisons are reliable only when they assess state‑space coverage and implement methods consistently, and it proposes simple checks for these conditions.
By Muhammad Ahmed Atif, Nehal Naeem Haji, Muhammad Ebad Atif, Mohammad Shahid Shaikh
The paper introduces P2NPCBench, a benchmark that pairs 12 NP‑complete problems with their polynomial‑time counterparts, using an adversarial multi‑agent system (AgentGAN) to rewrite problem statements until standard code models fail while preserving correctness and complexity class. The benchmark yields 2,114 hardened statements with verified human solutions, and evaluation shows that all tested models perform significantly worse on the NP‑complete versions (35.4% vs. 63.8% pass@1), with most failures due to incorrect or crashing programs rather than timeouts. The hardening process also reduces the performance gap between P and NP‑complete problems, indicating that prior gaps were partly due to familiarity with textbook phrasing.
By Nurul Labib Sayeedi, Md. Faiyaz Abdullah Sayeedi, Zabir Al Nazi, Mahbub E Sobhani, Shubhashis Roy Dipta
The paper investigates how mainstream document converters lose formatting cues—specifically highlights and colours—when turning Word or PDF files into tokens for enterprise agents. Using a dataset of 76 policy documents, the authors show that this loss dramatically degrades agents’ ability to recover author‑specified requirements, dropping micro‑F1 scores from 0.690 to as low as 0.075. The study also demonstrates that explicitly adding cues back into the text can partially recover performance and that the issue is measurable and inexpensive to fix.
By Yikai Zhao, Saurabh Pandey, Pradeep Kumar Misra
Environment Duels is a dynamic agentic benchmark that evaluates large language models (LLMs) on both creating interactive environments and solving them. In the study, nine frontier LLMs were tested, revealing that the best solvers are not necessarily the best authors, and that smaller models often produce the most challenging environments and useful hints. The benchmark also shows that pairwise interactions among models form a stable ranking, with independent authors converging on similar designs, and that a reinforcement‑learning learner’s improvement on another model’s environments correlates with its performance on held‑out reasoning and agentic tasks.
By Shibo Jin, Zhiqiu Xu, Yiling Zou, Shangyu Gong, Mayur Naik
The paper investigates how access to external tools affects the computational expressivity of finite‑precision recurrent models. It shows that any finite‑state bounded‑interface tool can be internalized with only a logarithmic increase in recurrent bits, while a single unbounded tape with simple read/write/move commands suffices for Turing completeness, enabling a single‑layer selective affine state‑space model to simulate any Turing machine with only logarithmic recurrent bits. The study provides exact separations between tool‑augmented and tool‑free regimes under deterministic semantics.
By Nikola Zubi\'c, Qian Li, Yuyi Wang, Davide Scaramuzza
HorizonComposer is a new instruction‑guided video editing framework that preserves traffic semantics in driving videos. It uses a hybrid dataset of real and synthetic edits, applies advantage conditioning with dense pixel‑level rewards during training, and introduces thinking frames to reduce temporal shock. Experiments show state‑of‑the‑art spatiotemporal consistency, improved content preservation, and gains in downstream road line segmentation and multi‑object detection.
By Mauricio Soroco, Yuqiu Liu, Zaid Tasneem, Francesco Pittaluga, Abhishek Aich, Wuyang Chen, Manmohan Chandraker, Ziyu Jiang
LocalAgent is a collaborative framework that combines a general multimodal large language model (MLLM) with lightweight specialist verifiers to verify fine‑grained instance‑level consistency between images. The MLLM first identifies suspicious local regions, which are then examined by the Local Evidence Verifier (LEV) toolkit—comprising Local Appearance Consistency (LAC) and Local Geometric Correspondence (LGC)—to provide structured scalar evidence for the final decision. Experiments demonstrate a 4.9% accuracy gain over the strongest baseline while invoking the verifier on only 12.13% of samples, preserving the MLLM’s general multimodal capabilities.
By Jiachen Guo, Xinshan Zhu, Boyun Wang, Yanyan Liang, Jun Wan, Feng Lin, Ajian Liu
DetM-Tracker is a lightweight Video Instance Segmentation framework designed for autonomous industrial inspection. It reduces memory and computation by condensing instance features into four-scalar descriptors, dynamically allocating query capacity, and using entropy-gated skip connections to focus processing on uncertain regions. The method achieves roughly 20× lower inference latency and 6× less VRAM usage than dense-feature baselines.
By You-Rui Lin, Jen-Chieh Wu, Tzu-Ti Wei, Jen-Jee Chen, Yu-Chee Tseng
The paper introduces SearchWeaver, a framework that automatically optimizes search interfaces for language-model agents by revising corpus artifacts, executable tools, and usage instructions based on execution trajectories and task rewards. By tailoring the interface to each corpus, the method improves answer accuracy from 71.6% to 78.4% on BrowseComp-Plus and raises average nDCG@10 from 45.7 to 55.0 across four BRIGHT subsets. These results demonstrate the benefit of jointly optimizing interface components to enhance agentic search performance.
By Sangjune An, Jaeyoon Jung, Yumin Choi, Jinheon Baek, Junmo Cho, Sung Ju Hwang, Hae Beom Lee
The paper surveys emerging video understanding agents that use video as the primary source and actively decide what information to inspect, retain, verify, and act upon. It formalizes an agent loop for video understanding and organizes the literature through a challenge‑to‑design taxonomy linking context bottlenecks, evidence sparsity, temporal causality, and multimodal ambiguity to corresponding design solutions. The survey also reviews state‑space paradigms, learning paradigms, supervision signals, benchmarks, evaluation protocols, and outlines open directions toward agentic‑native temporal modeling and video‑native agents.
By Xinyu Deng, Siwen Luo, Daochang Liu
The paper introduces a new methodology and analytical framework to diagnose errors in the intermediate execution traces of autonomous Deep Search and Deep Research Large Language Model agents. It presents a taxonomy that classifies these errors into System‑Level and Task‑Level categories, and uses systematic examination of step‑by‑step traces to show how they impact overall performance. The authors also provide empirical experiments and design guidelines aimed at improving the robustness, reasoning, and reliability of these agents.
By Zixuan Yu, Zhenheng Tang, Xiaowen Chu, Bo Han
The paper investigates automated failure attribution (AFA) in large language model (LLM)-based agentic systems, where multiple agents collaborate on complex tasks. It finds that LLMs contain sufficient internal knowledge to attribute failures but often fail to output the correct step, revealing a knows‑but‑won’t‑say gap. To address this, the authors introduce grafting—a manual intervention that leverages latent model knowledge—showing it consistently outperforms direct prompting across several LLMs and benchmarks.
By Wendy Zheng, Liang Wu, Yinhan He, Mayank Darbari, Jundong Li, Liangjie Hong
ExpEvolve introduces idea-level extraction and control into evolutionary coding agents, enabling the discovery of high-level factors that explain differences in program performance. By leveraging these factors, the framework guides both exploitation of existing experience and exploration when progress stalls, leading to significant gains in sample efficiency and final performance. Across 11 open-ended mathematical tasks and additional benchmarks, ExpEvolve improves results by up to 5.0× and transfers benefits to other large language models and machine learning tasks.
By Yongqiang Chen, Chenxi Liu, Deyu Zou, Zhenhao Chen, Tongliang Liu, Bo Han, Kun Zhang
The paper introduces Contrastive Attribution Topology Search (CATS), a method for selecting compact, high‑performing teams of agents within a larger multi‑agent system. CATS uses an LLM attributer to assign credit to agents based on execution traces, then iteratively prunes weak edges and agents to form a small, efficient team. Experiments show CATS improves accuracy on BigCodeBench by 5.3 percentage points over heuristic search while reducing deployment tokens by 54%, matches strong baselines on MBPP with fewer tokens, and successfully removes adversarial agents in a HotpotQA stress test.
By Fabian Tritthart, Tom Zehle
arXiv:2610.29671v1 Announce Type: new
Abstract: Long-term agent memory is shifting from predefined workflows toward learnable policies for memory construction and retrieval. Existing learning-based m...
By Zhiqian Xia, Jielong Tang, Ling-An Zeng, Shixiong Kai, Rong Yao, Hao Kong, Xiao Dong, Wei Wang, Mingxuan Yuan, Siyu Xia, Haifeng Xia
arXiv:2610.29408v1 Announce Type: new
Abstract: Interactive video world models predict how an environment evolves under user actions, but most existing systems assume one controlled agent and one tas...
By Xinyu Wang, Huafeng Shi, Yue Ma, Zijie Meng, Teng Xi, Yuzhuo Chen, Yucheng Wang, Ruicheng Zhang, Guoxin Zhang, Pengfei Wan, Xiu Li, Fangneng Zhan
arXiv:2610.29936v1 Announce Type: new
Abstract: Recovering metric world-space hand motion from egocentric video supports understanding human interactions and transferring them to embodied agents. Thr...
By Naru Suzuki, Satoshi Ikehata, Takehiko Ohkawa, Yoichi Sato, Shuhei Kurita
arXiv:2603.17068v3 Announce Type: replace
Abstract: Structured 3D representations such as keypoints and meshes offer compact, expressive descriptions of deformable objects, jointly capturing geometri...
By Yeheng Zong, Yizhou Chen, Alexander Bowler, Chia-Tung Yang, Ram Vasudevan
arXiv:2610.05195v3 Announce Type: replace
Abstract: The next generation of multimodal research agents must reason over long-lived research histories rather than short model completions. During a sing...
By On Tai Tang, Huanyao Zhang, Jiepeng Zhou, Wentao Zhang
CineEvolve is a new agentic self‑evolution framework for cinematic multi‑shot video generation that integrates a Director and a Critic into a single vision‑language policy. The system alternates reinforcement learning steps, letting the Director identify flaws and the Critic refine its guidance, while a novel Evolutionary GRPO (E‑GRPO) stabilizes co‑evolution by anchoring history and safeguarding revisions. The authors also release CineTraj, a 40K‑sample dataset of cinematic trajectories, and show that CineEvolve outperforms existing methods in extensive experiments.
By Jingzhe Zhang, Peixuan Zhang, Genying Li, Bing Ma, Shuchen Weng, Si Li, Boxin Shi