Contrastive On-Policy Distillation (COPD) is a framework that improves on-policy distillation by using a frozen teacher to evaluate student states under two contrasting prompts—one encouraging low reasoning effort and one encouraging high effort. The difference in log‑probabilities between these prompts provides a token‑level advantage signal that guides the student toward more concise and efficient reasoning strategies. Experiments on nine multimodal benchmarks show that COPD reduces reasoning length while maintaining task performance, and the contrastive approach can also be applied to on‑policy self‑distillation, allowing a model to compress its own reasoning without an external teacher.
By Jiacheng Ruan, Jun Tang, Wenzhen Yuan, Ting Liu, Shuai Bai, Dayiheng Liu, Zhibo Yang, Yuzhuo Fu
FoCLIP is a framework that creates adversarial examples to manipulate CLIP-based image quality metrics by reducing the alignment between image and text features. It uses stochastic gradient descent to combine feature alignment, score distribution balancing, and pixel‑guard regularization, enabling high CLIPscore predictions while maintaining visual fidelity. Experiments on artistic prompts and ImageNet show significant CLIPscore gains, and the authors also propose a color‑channel sensitivity detection method that achieves 91% accuracy.
By Yulin Chen, Zeyuan Wang, Tianyuan Yu, Yingmei Wei, Liang Bai
Efficient Multi-Modal Planning with Reward-Guided Preference Optimization for Autonomous Driving proposes EMPlan, a hybrid trajectory planning method that combines sparse anchors with an offset refinement module for low-latency, high-accuracy predictions. The approach uses a two-stage training paradigm—pretraining followed by reward-guided fine-tuning—to improve safety without extra inference cost, leveraging rule-based reward signals and unpaired preference supervision. EMPlan is evaluated on the non-reactive NAVSIM benchmark, achieving a favorable balance between planning accuracy and efficiency under real-time constraints.
By Chenglin Chen, Lujia Wang, Xinhu Zheng, Jun Ma, Haoang Li
The paper introduces Experimental Experience Modeling (EEM), a framework that helps autonomous research agents decide which experiments to run by reusing and accumulating past experimental data. EEM extracts decision-relevant records from previous experiments, stores them in an experience library, and retrieves relevant information when making new experimental choices. If historical data is insufficient, EEM performs a low-cost pilot experiment to gather the needed experience before committing to a full-scale, resource-intensive study, thereby iteratively expanding the library and improving research efficiency.
By Wenda Wei, Yingchen Zhang, Ruqing Zhang, Jiafeng Guo, Daiting Shi, Xueqi Cheng
OSWorld-Science is a benchmark and evaluation environment for computer-using agents that use visual language models (VLMs) to perform scientific software tasks. It includes 12 VLMs and 146 high-quality tasks across domains such as molecular drawing, pathology image analysis, statistical computing, and physical simulation, with artifact-based evaluation and a harness that logs interactions and supports model comparison. The benchmark was developed through expert proposals and iterative human–AI co‑design, and results show that current VLMs still struggle with key scientific questions, offering insights into factors like language, reasoning, and context length.
By Dingyuan Dai, Heli Qi, Lei Liu, Yinxi Li, Baiding Chen, Zijun Dou, Qingcheng Zeng, Qi Kang, Oliver Sun, Eric Wang, Bo Zhou, Haixin Wang, Yufan Du, Shi Bo, Ruihan Lin, Mengqi Yuan, Dunjie Lu, Steven Dillmann, Yiming Shi, Tina Su, Amy Xin, Minghao Liu, Xi Wang, Xu Huang, Ge Zhang, Pengyu Nie, Zhen Yang, Jie Tang, Juanzi Li, Weihao Xuan, Tianyu Liu
E2E-SWE is a benchmark that tests large language models’ ability to create complete, functional software repositories from scratch. It includes 186 tasks across 11 programming languages, each requiring an agent to build an installable project based solely on a natural‑language specification and an empty workspace, while passing a hidden test suite. The benchmark was crafted by software engineers and LLMs, then refined through iterative verification by autonomous agents to ensure clarity and solvability.
By Hantian Ding, Chloe Bi, Jiacheng Zhu, John Yang, Matt Deitke, Pengcheng Yin, Zijian Wang, Rui Hou
EviRover is a perception agent that goes beyond a single glance by actively gathering information to resolve perceptual queries. The authors created two data generation pipelines, producing EviRover-SFT-5K and EviRover-RL-12K, and a human‑verified benchmark called EviLens with 688 instances across five perception categories. Trained with supervised fine‑tuning and agentic reinforcement learning, the 4B EviRover outperforms its backbone by an average of 30 points on EviLens and shows strong transfer to other benchmarks such as WebEyes and BrowseComp‑VL.
By Kaixuan Fan, Kaituo Feng, Tianshuo Peng, Yilei Jiang, Manyuan Zhang, Junke Wang, Xiangyu Yue
The paper presents a large‑scale, 3GPP TR 38.901‑compliant dataset for RIS‑aided millimeter‑wave B5G networks, covering 20 deployment variants with diverse user densities, fading, and blockage conditions. Each sample includes oracle RIS phase configurations from a brute‑force search, along with full CSI, per‑link channel decomposition, optimal phase matrices, and CQI labels, enabling a wide range of machine‑learning tasks. The authors also introduce a novel CSI‑to‑CQI mapping as a benchmark for scalable link‑quality prediction and evaluate it against state‑of‑the‑art models under various conditions.
By Pujitha Mamillapalli, Pankaj Singh Rathour, Abhinav Kumar
The paper introduces LogiC-Diff, a logic-conditioned bi-stage diffusion framework that embeds Signal Temporal Logic (STL) specifications into AI-enabled cyber‑physical system (CPS) forecasting models. By using STL as a conditioning signal, the method repairs inputs and refines outputs to jointly mitigate adversarial perturbations and enforce desired temporal behaviors. Experiments on two real‑world CPS datasets show that LogiC-Diff consistently improves robustness and specification compliance across various sensor faults and cyber attacks, outperforming reconstruction‑based defenses.
By Ziyan An, John Stankovic, Meiyi Ma
DiffWAM is a geometry‑conditioned navigation world‑action model that transforms predictive features from a frozen video foundation model into continuous camera trajectories, eliminating the need for future‑video synthesis and multi‑frame reconstruction during deployment. Its Grid‑Motion module preserves spatial‑temporal motion associations, while Latent2Pose grounds them with first‑frame geometry to recover metrically meaningful 3D motion. The system, complemented by FastDreamer for asynchronous trajectory handoff, achieves a trajectory RMSE of 0.3492 m and a 74.40 % endpoint success rate on the DiffWAM‑1000 benchmark, with real‑world tests showing complex UAV behaviors and an onboard implementation reaching 1.08 s latency on NVIDIA Jetson AGX Thor.
By Mo Zhu, Yuze Wu, Xijie Huang, Xiao Cui, Fei Gao, Xin Zhou
The paper introduces a new hyperspectral image dataset for benchmarking salient object detection, comprising 60 hyperspectral images, their ground‑truth binary masks, and corresponding sRGB renderings. The dataset was curated to include diverse object sizes, counts, contrasts, and positions, addressing the lack of dedicated hyperspectral data for this task. The authors also evaluate existing hyperspectral saliency models using the AUC metric and provide the dataset on GitHub and Hugging Face.
By Nevrez Imamoglu, Yu Oishi, Xiaoqiang Zhang, Guanqun Ding, Yuming Fang, Toru Kouyama, Ryosuke Nakamura
GATE-ST is a gene-aware text-image encoder that enhances spatial transcriptomics predictions by integrating gene descriptions into image-based models. The method encodes gene summaries with a text encoder and fuses these embeddings with image features via cross‑attention, aligning them with morphological cues. Benchmarks show GATE‑ST outperforms random gene embeddings and other image‑text fusion architectures, indicating its potential to improve accuracy while reducing time and cost in spatial gene expression analysis.
By Lucas Ni, Jian Luo, Wentao Huang, Chao Chen
The paper introduces Belief-Aware Multi-Agent Path Finding under Map Uncertainty, addressing the challenge that real-world environments can change unexpectedly. It proposes MAGIC, a framework that uses a Gaussian Markov Random Field and Gaussian Belief Propagation to update a shared belief about traversability online, allowing agents to infer the state of nearby unobserved locations. Experiments on standard MAPF benchmarks show that MAGIC reduces the executed sum of costs on 96.3% of instances, outperforming existing approaches across various planner families and large agent teams.
By Viraj Parimi, Shao-Hung Chan, Han Zhang, Jingkai Chen, Brian Williams
The paper introduces BACKDROP, a benchmark that evaluates how well AI agents maintain their capabilities when faced with everyday hazards in dynamic environments. BACKDROP adds four types of hazards—authority, injection, boundary, and fault—to a task’s execution environment and measures whether agents can still achieve the correct end state. Across 3,678 variants and 16 models, the average success rate drops dramatically from 69.5% to 31.3% when all hazards are present, revealing that agents often follow unauthorized requests and fail to resist injected text.
By Nusrat Jahan Lia, Shubhashis Roy Dipta
Steering Fields introduce adaptive vector fields that re-estimate steering directions at each step of a flow-based text-to-image generation process, replacing the fixed global steering vectors traditionally used. By operating on noisy states, they provide a continuous trade-off between steering strength and content preservation, and allow simultaneous induction and inhibition of concepts without explicit spatial masks or object priors. The method achieves state-of-the-art safety steering benchmarks and can also function as a structure-preserving image-editing technique, delivering high semantic fidelity while remaining model-agnostic and inversion-free.
By Simone Facchiano, Jan Eric Lenssen, Bernt Schiele, Wolfgang Stammer, Fabio Galasso, Jonas Fischer
arXiv:2609.38362v1 Announce Type: new
Abstract: Generative vision-language models (VLMs) such as Qwen-VL and LLaVA achieve strong zero-shot performance on tasks overlapping with their pretraining dis...
By Hung-Jen Chen, Yu-Heng Ho, Ting-Yao Huang, Po-Hsiang Hsu, Li-Yu Chen, Chun-Yi Lee, Min Sun
arXiv:2609.38864v1 Announce Type: new
Abstract: Embodied tasks demand accurate, flexible, and semantically rich 3D scene representations. 3D semantic occupancy is well suited to this requirement, as...
By Jinglong Wang, Yunjie Wang, Zhiyang Zhang, Jiawei He, Ye Yuan, Bo Qiu, Jing Zhang
arXiv:2609.39265v1 Announce Type: new
Abstract: The Segment Anything Model (SAM) achieves remarkable performance in visual segmentation. The latest SAM3 extends promptable segmentation to concept-lev...
By Ziqi Zhou, Yifan Hu, Yufei Song, Haowen Jiang, Xianlong Wang, Shengshan Hu, Dezhong Yao, Leo Yu Zhang
arXiv:2609.39527v1 Announce Type: new
Abstract: The presence of lens flares in images can significantly reduce the quality of downstream application results for tasks such as 3D scene reconstruction....
By Tarun Yenamandra, Jonathon Luiten, Daniel Cremers, Nathan Matsuda
arXiv:2609.40358v1 Announce Type: new
Abstract: Video world models are expected to predict how the physical world evolves, yet they often produce visually plausible videos that violate basic physical...
By Liming Lu, Xianzheng Ma, Wenkun He, Guanqi Zhan, Yilin Zhao, Junyu Chen, Mengyao Xu, Jiaojiao Fan, Wenhang Ge, Yuchao Gu, Yunze Liu, Boyi Li, Zhen Dong, Victor Prisacariu, Ming-Yu Liu, Song Han, Han Cai