The paper introduces EXPO, an algorithm that automatically optimizes the meta-prompt—specifically the task description and meta-instruction—for large language model agents in sequential decision-making tasks such as Bayesian optimization and multi-armed bandits. Building on adversarial bandit techniques to handle non-stationary rewards, the authors extend EXPO to EXPO-ES, which also optimizes exemplars (historical interactions) within the meta-prompt. Experiments demonstrate that these methods significantly improve the performance of LLM-based agents in sequential decision-making scenarios.
By Mingze Kong, Zhiyong Wang, Yao Shu, Zhongxiang Dai
PRISM is an agentic retrieval framework that uses large language models in a structured loop to improve evidence gathering for multi‑hop question answering. It splits retrieval into three specialized agents—a Question Analyzer, a Selector focused on precision, and an Adder focused on recall—whose iterative interaction yields a compact yet comprehensive evidence set. Experiments on HotpotQA, 2WikiMultiHopQA, MuSiQue, and MultiHopRAG show that PRISM consistently outperforms strong baselines by achieving higher retrieval accuracy and filtering out distracting content.
By Md Mahadi Hasan Nahid, Davood Rafiei
The paper introduces Cognitive Chain-of-Thought (CoCoT), a structured reasoning framework for vision‑language models that divides multimodal social reasoning into three cognitively inspired stages: Perception, Situation, and Norm. CoCoT improves performance across diverse tasks—multimodal intent disambiguation, theory of mind, social commonsense reasoning, and safety instruction following—by 5.9% to 4.6% on average. Fine‑tuning on CoCoT‑structured traces further boosts accuracy by 5–6% without explicit prompting, indicating that models internalize the structured reasoning pattern and that the approach enhances interpretability and social alignment in multimodal systems.
By Eunkyu Park, Wesley Hanwen Deng, Gunhee Kim, Motahhare Eslami, Maarten Sap
arXiv:2608.27584v1 Announce Type: new
Abstract: Autonomous systems rely on extracting information from light, yet remain brittle in extreme environments, from nighttime navigation to high-speed robot...
By Varun Sundar, Pavan Thodima, Sacha Jungerman, Mohit Gupta
arXiv:2608.27997v1 Announce Type: new
Abstract: Air-ground cross-view referring person detection is a necessary component in the language-to-perception-to-control chain of collective embodied intelli...
By Zhoupeng Guo, Xinjie Yao, Yunqi Zhu, Zhihe Fan, Siqi Zhao, Jianjun Chen, Yichen Dong, Yan Fan, Pengfei Zhu
arXiv:2608.27871v1 Announce Type: new
Abstract: Long-video understanding remains challenging for Multimodal Large Language Models (MLLMs) due to limited context length. Uniform sampling may miss cruc...
By Ziling Huang, Shin'ichi Satoh
arXiv:2606.17092v2 Announce Type: replace-cross
Abstract: Agentic systems are increasingly integrated with geographic information systems (GIS), where multi-agent coordination enables complex convers...
By Kyle Gao, Pranavi Kotta, Linlin Xu, Jonathan Li, David A. Clausi
D3‑Gym is the first automatically constructed dataset that provides verifiable environments for scientific data‑driven discovery, comprising 565 tasks from 239 real scientific repositories across four disciplines. Each task includes a natural‑language instruction, an executable environment with pre‑installed dependencies, dataset previews, a reference solution, and an automatically synthesized evaluation script that achieves 87.5% agreement with human‑annotated gold standards. Training on trajectories sampled from D3‑Gym consistently improves Qwen3 models on ScienceAgentBench, and the platform also serves as a testbed for studying agentic optimization loops such as Autoresearch on real scientific workflows.
By Hanane Nour Moussa, Yifei Li, Zhuoyang Li, Yankai Yang, Cheng Tang, Tianshu Zhang, Nesreen K. Ahmed, Ali Payani, Ziru Chen, Huan Sun
The paper investigates how scaling video diffusion models affects performance on autonomous driving data, training models ranging from 1 M to 9 B parameters on up to 5,500 hours of driving footage. Validation loss follows predictable power‑law trends with both model size and training exposure, showing that longer training yields faster loss reduction than increasing model size, though larger models still achieve lower asymptotic loss. Guided by these findings, the authors train a 9 B‑parameter model that becomes the largest video diffusion model trained from scratch on driving data and sets a new open‑source state‑of‑the‑art on the nuScenes benchmark.
By Victor Besnier, Anh-Quan Cao, Elias Ramzi, Spyros Gidaris, Tuan-Hung Vu, Andrei Bursuc, Eloi Zablocki, Matthieu Cord
The paper proposes a label‑free 3D perception framework where roadside units (RSUs) act as unsupervised teachers for self‑driving cars. RSUs learn local 3D detectors from unlabeled data and broadcast predictions to passing vehicles, which use these as pseudo‑labels to train an ego‑centric detector. In a CARLA simulation, the method achieves 82.3% AP for vehicle detection, approaching a fully supervised upper bound of 94.4%, and demonstrates scalability and complementarity with existing ego‑centric approaches.
By Zhen Xu, Jinsu Yoo, Cristian Bautista, Zanming Huang, Tai-Yu Pan, Zhenzhen Liu, Katie Z Luo, Mark Campbell, Bharath Hariharan, Wei-Lun Chao
The paper introduces VICT, a method that leverages the internal structure of verifiable tasks to perform fine‑grained credit assignment for long‑horizon LLM agents. VICT exposes executable or evidence‑backed atoms from a task’s terminal verifier and traces them back to actions via dependency‑valid proof edges, redistributing advantage only along these edges. This approach improves performance on ALFWorld and WebShop compared to outcome‑only training and matches recent fine‑grained credit methods without requiring additional critics, labels, or inference‑time verifier access.
By Pengcheng Li, Zhengyang Zhang, Dongxu Zhang, Sui Huang, Shaohua Ma
arXiv:2601.13247v2 Announce Type: replace-cross
Abstract: Current Large Language Models (LLMs) exhibit a critical modal disconnect: they possess vast semantic knowledge but lack the procedural ground...
By Baochang Ren, Yunzhi Yao, Rui Sun, Shuofei Qiao, Ningyu Zhang, Huajun Chen
arXiv:2608.27549v1 Announce Type: new
Abstract: Physical understanding and reasoning depend on forming compact and generalizable representations of the world. While modern vision-language models can...
By Hanyang Wang, Yimo Cai, Weiliang Chen, Jiawei Chi, Haowen Sun, Qiyu Dai, Yi-Hsin Hung, Xingzhuo Guo, Jinshan Ren, Runmao Yao, Ziwei Liu, Mingsheng Long, Yueqi Duan, Jun Gao, Jiangran Lyu, Fangfu Liu, Jialong Wu
arXiv:2608.27505v1 Announce Type: new
Abstract: Reinforcement learning from human feedback (RLHF) has become the dominant paradigm for aligning large language models (LLMs) with human preferences. Ho...
By Zifei Shan, Fangning Shao
arXiv:2605.12015v3 Announce Type: replace-cross
Abstract: Reusable skills are becoming a common interface for extending large language model agents, packaging procedural guidance with access to files...
By Chang Jin, An Wang, Zeming Wei, Kai Wang, Biaojie Zeng, Qiaosheng Zhang, Chao Yang, Jingjing Qu, Xia Hu, Xingcheng Xu
arXiv:2608.28198v1 Announce Type: new
Abstract: Privacy-preserving learning is often motivated by the idea that protecting users' data can preserve trust and thus participation, improving utility in...
By Uddalak Mukherjee, Edwige Cyffers, Yann Chevaleyre
ContextPilot is a proactive context‑management framework designed to improve long‑horizon agentic reasoning with large language models. It expands the toolset to include planning, long‑term memory, and soft context offloading, and introduces a reinforcement‑learning strategy that focuses on critical editing decisions and assigns action‑level advantages. Experiments on long‑context QA and deep search tasks demonstrate that ContextPilot achieves stronger performance with a more compact working context, outperforming existing baselines across various base models and benchmarks.
By Zhuoshi Pan, Qizhi Pei, Junru Lu, Honglin Lin, H. Vicky Zhao, Di Yin, Xing Sun
The paper reports that a model can pass fidelity checks—verifying that extracted values match the source—without actually opening a datasheet, due to a hidden constraint that disables tool use. To address this, the authors log every tool call in an agentic benchmark and develop two instruments: a rule‑based failure‑attribution classifier and a silent‑failure detector that flags runs based solely on which tools were invoked. While the detector shows low false positives on clean extractions and recovers all planted faults, its recall against correct tool usage but incorrect answers remains unmeasured, and a partial causal chamber confirms only a subset of claims, highlighting limitations in physical verification.
By Qing Ye, Meng-Hsuan Lin
RealClawBench is a live benchmark framework derived from real OpenClaw developer‑agent sessions, designed to capture the distribution, diversity, and real‑world difficulty of deployed agent use. It reconstructs execution environments and uses deterministic verifiable scorers to convert real sessions into reproducible, automatically scored tasks, yielding 281 executable tasks with minimal distribution shift. Evaluation of 14 contemporary models shows the best system solves only 65.8% of tasks, highlighting significant room for improvement on realistic workloads.
By Zongwei Lv, Yaoming Li, Zhewen Tan, Yilun Yao, Yuxuan Tian, Lin Sun, Xiangzheng Zhang, Weihong Lin, Tong Yang, Guangxiang Zhao
ZipMVS is a multi‑view stereo (MVS) approach that focuses on reducing memory usage while maintaining high‑quality 3D reconstructions. It introduces a new depth‑hypothesis strategy that compresses the cost volume, significantly lowering GPU memory consumption. Experiments on the DTU and Tanks and Temples datasets demonstrate that ZipMVS delivers competitive reconstruction quality compared to other efficiency‑oriented MVS methods.
By Guanglin Jin, Hongshan Yu, Javier Civera, Zhaoxin Li