Yi Wan

Youjie Zhou, Guofeng Mei, Yiming Wang et al.

This paper presents an investigation into the estimation of optical and scene flow using RGBD information in scenarios where the RGB modality is affected by noise or captured in dark environments. Existing methods typically rely solely on RGB images or fuse the modalities at later stages, which can result in lower accuracy when the RGB information is unreliable. To address this issue, we propose a novel deep neural network approach named FusionRAFT, which enables early-stage information fusion between sensor modalities (RGB and depth). Our approach incorporates self- and cross-attention layers at different network levels to construct informative features that leverage the strengths of both modalities. Through comparative experiments, we demonstrate that our approach outperforms recent methods in terms of performance on the synthetic dataset Flyingthings3D, as well as the generalization on the real-world dataset KITTI. We illustrate that our approach exhibits improved robustness in the presence of noise and low-lighting conditions that affect the RGB images. We release the code, models and dataset at https://github.com/jiesico/FusionRAFT.

Yi Wan, Richard S. Sutton

We show two average-reward off-policy control algorithms, Differential Q-learning (Wan, Naik, & Sutton 2021a) and RVI Q-learning (Abounadi Bertsekas & Borkar 2001), converge in weakly communicating MDPs. Weakly communicating MDPs are the most general MDPs that can be solved by a learning algorithm with a single stream of experience. The original convergence proofs of the two algorithms require that the solution set of the average-reward optimality equation only has one degree of freedom, which is not necessarily true for weakly communicating MDPs. To the best of our knowledge, our results are the first showing average-reward off-policy control algorithms converge in weakly communicating MDPs. As a direct extension, we show that average-reward options algorithms for temporal abstraction introduced by Wan, Naik, & Sutton (2021b) converge if the Semi-MDP induced by options is weakly communicating.

Yi Wan, Richard S. Sutton

We propose a new objective for option discovery that emphasizes the computational advantage of using options in planning. In a sequential machine, the speed of planning is proportional to the number of elementary operations used to achieve a good policy. For episodic tasks, the number of elementary operations depends on the number of options composed by the policy in an episode and the number of options being considered at each decision point. To reduce the amount of computation in planning, for a given set of episodic tasks and a given number of options, our objective prefers options with which it is possible to achieve a high return by composing few options, and also prefers a smaller set of options to choose from at each decision point. We develop an algorithm that optimizes the proposed objective. In a variant of the classic four-room domain, we show that 1) a higher objective value is typically associated with fewer number of elementary planning operations used by the option-value iteration algorithm to obtain a near-optimal value function, 2) our algorithm achieves an objective value that matches it achieved by two human-designed options 3) the amount of computation used by option-value iteration with options discovered by our algorithm matches it with the human-designed options, 4) the options produced by our algorithm also make intuitive sense--they seem to move to and terminate at the entrances of rooms.

Peihao Wu, Yongxiang Yao, Yi Wan et al.

Synthetic Aperture Radar (SAR) and optical imagery provide complementary strengths that constitute the critical foundation for transcending single-modality constraints and facilitating cross-modal collaborative processing and intelligent interpretation. However, existing benchmark datasets often suffer from limitations such as single spatial resolution, insufficient data scale, and low alignment accuracy, making them inadequate for supporting the training and generalization of multi-scale foundation models. To address these challenges, we introduce SOMA-1M (SAR-Optical Multi-resolution Alignment), a pixel-level precisely aligned dataset containing over 1.3 million pairs of georeferenced images with a specification of 512 x 512 pixels. This dataset integrates imagery from Sentinel-1, PIESAT-1, Capella Space, and Google Earth, achieving global multi-scale coverage from 0.5 m to 10 m. It encompasses 12 typical land cover categories, effectively ensuring scene diversity and complexity. To address multimodal projection deformation and massive data registration, we designed a rigorous coarse-to-fine image matching framework ensuring pixel-level alignment. Based on this dataset, we established comprehensive evaluation benchmarks for four hierarchical vision tasks, including image matching, image fusion, SAR-assisted cloud removal, and cross-modal translation, involving over 30 mainstream algorithms. Experimental results demonstrate that supervised training on SOMA-1M significantly enhances performance across all tasks. Notably, multimodal remote sensing image (MRSI) matching performance achieves current state-of-the-art (SOTA) levels. SOMA-1M serves as a foundational resource for robust multimodal algorithms and remote sensing foundation models. The dataset will be released publicly at: https://github.com/PeihaoWu/SOMA-1M.

Yi Wan, Ali Rahimi-Kalahroudi, Janarthanan Rajendran et al.

In recent years, a growing number of deep model-based reinforcement learning (RL) methods have been introduced. The interest in deep model-based RL is not surprising, given its many potential benefits, such as higher sample efficiency and the potential for fast adaption to changes in the environment. However, we demonstrate, using an improved version of the recently introduced Local Change Adaptation (LoCA) setup, that well-known model-based methods such as PlaNet and DreamerV2 perform poorly in their ability to adapt to local environmental changes. Combined with prior work that made a similar observation about the other popular model-based method, MuZero, a trend appears to emerge, suggesting that current deep model-based methods have serious limitations. We dive deeper into the causes of this poor performance, by identifying elements that hurt adaptive behavior and linking these to underlying techniques frequently used in deep model-based RL. We empirically validate these insights in the case of linear function approximation by demonstrating that a modified version of linear Dyna achieves effective adaptation to local changes. Furthermore, we provide detailed insights into the challenges of building an adaptive nonlinear model-based method, by experimenting with a nonlinear version of Dyna.

Runyu Jiao, Yi Wan, Fabio Poiesi et al.

The increasing popularity of compact and inexpensive cameras, e.g.~dash cameras, body cameras, and cameras equipped on robots, has sparked a growing interest in detecting anomalies within dynamic scenes recorded by moving cameras. However, existing reviews primarily concentrate on Video Anomaly Detection (VAD) methods assuming static cameras. The VAD literature with moving cameras remains fragmented, lacking comprehensive reviews to date. To address this gap, we endeavor to present the first comprehensive survey on Moving Camera Video Anomaly Detection (MC-VAD). We delve into the research papers related to MC-VAD, critically assessing their limitations and highlighting associated challenges. Our exploration encompasses three application domains: security, urban transportation, and marine environments, which in turn cover six specific tasks. We compile an extensive list of 25 publicly-available datasets spanning four distinct environments: underwater, water surface, ground, and aerial. We summarize the types of anomalies these datasets correspond to or contain, and present five main categories of approaches for detecting such anomalies. Lastly, we identify future research directions and discuss novel contributions that could advance the field of MC-VAD. With this survey, we aim to offer a valuable reference for researchers and practitioners striving to develop and advance state-of-the-art MC-VAD methods.

Yi Wan, Huizhen Yu, Richard S. Sutton

This paper analyzes reinforcement learning (RL) algorithms for Markov decision processes (MDPs) under the average-reward criterion. We focus on Q-learning algorithms based on relative value iteration (RVI), which are model-free stochastic analogues of the classical RVI method for average-reward MDPs. These algorithms have low per-iteration complexity, making them well-suited for large state space problems. We extend the almost-sure convergence analysis of RVI Q-learning algorithms developed by Abounadi, Bertsekas, and Borkar (2001) from unichain to weakly communicating MDPs. This extension is important both practically and theoretically: weakly communicating MDPs cover a much broader range of applications compared to unichain MDPs, and their optimality equations have a richer solution structure (with multiple degrees of freedom), introducing additional complexity in proving algorithmic convergence. We also characterize the sets to which RVI Q-learning algorithms converge, showing that they are compact, connected, potentially nonconvex, and comprised of solutions to the average-reward optimality equation, with exactly one less degree of freedom than the general solution set of this equation. Furthermore, we extend our analysis to two RVI-based hierarchical average-reward RL algorithms using the options framework, proving their almost-sure convergence and characterizing their sets of convergence under the assumption that the underlying semi-Markov decision process is weakly communicating.

Zheqing Zhu, Rodrigo de Salvo Braz, Jalaj Bhandari et al.

Reinforcement learning (RL) is a versatile framework for optimizing long-term goals. Although many real-world problems can be formalized with RL, learning and deploying a performant RL policy requires a system designed to address several important challenges, including the exploration-exploitation dilemma, partial observability, dynamic action spaces, and safety concerns. While the importance of these challenges has been well recognized, existing open-source RL libraries do not explicitly address them. This paper introduces Pearl, a Production-Ready RL software package designed to embrace these challenges in a modular way. In addition to presenting benchmarking results, we also highlight examples of Pearl's ongoing industry adoption to demonstrate its advantages for production use cases. Pearl is open sourced on GitHub at github.com/facebookresearch/pearl and its official website is pearlagent.github.io.

Wenfei Zhang, Ruipeng Zhao, Yongxiang Yao et al.

Synthetic Aperture Radar (SAR) and optical image registration is essential for remote sensing data fusion, with applications in military reconnaissance, environmental monitoring, and disaster management. However, challenges arise from differences in imaging mechanisms, geometric distortions, and radiometric properties between SAR and optical images. As image resolution increases, fine SAR textures become more significant, leading to alignment issues and 3D spatial discrepancies. Two major gaps exist: the lack of a publicly available multi-resolution, multi-scene registration dataset and the absence of systematic analysis of current methods. To address this, the MultiResSAR dataset was created, containing over 10k pairs of multi-source, multi-resolution, and multi-scene SAR and optical images. Sixteen state-of-the-art algorithms were tested. Results show no algorithm achieves 100% success, and performance decreases as resolution increases, with most failing on sub-meter data. XoFTR performs best among deep learning methods (40.58%), while RIFT performs best among traditional methods (66.51%). Future research should focus on noise suppression, 3D geometric fusion, cross-view transformation modeling, and deep learning optimization for robust registration of high-resolution SAR and optical images. The dataset is available at https://github.com/betterlll/Multi-Resolution-SAR-dataset-.

Peihao Wu, Yongxiang Yao, Wenfei Zhang et al.

Multimodal remote sensing image (MRSI) matching is pivotal for cross-modal fusion, localization, and object detection, but it faces severe challenges due to geometric, radiometric, and viewpoint discrepancies across imaging modalities. Existing unimodal datasets lack scale and diversity, limiting deep learning solutions. This paper proposes MapGlue, a universal MRSI matching framework, and MapData, a large-scale multimodal dataset addressing these gaps. Our contributions are twofold. MapData, a globally diverse dataset spanning 233 sampling points, offers original images (7,000x5,000 to 20,000x15,000 pixels). After rigorous cleaning, it provides 121,781 aligned electronic map-visible image pairs (512x512 pixels) with hybrid manual-automated ground truth, addressing the scarcity of scalable multimodal benchmarks. MapGlue integrates semantic context with a dual graph-guided mechanism to extract cross-modal invariant features. This structure enables global-to-local interaction, enhancing descriptor robustness against modality-specific distortions. Extensive evaluations on MapData and five public datasets demonstrate MapGlue's superiority in matching accuracy under complex conditions, outperforming state-of-the-art methods. Notably, MapGlue generalizes effectively to unseen modalities without retraining, highlighting its adaptability. This work addresses longstanding challenges in MRSI matching by combining scalable dataset construction with a robust, semantics-driven framework. Furthermore, MapGlue shows strong generalization capabilities on other modality matching tasks for which it was not specifically trained. The dataset and code are available at https://github.com/PeihaoWu/MapGlue.

Huizhen Yu, Yi Wan, Richard S. Sutton

This paper investigates the stability and convergence properties of asynchronous stochastic approximation (SA) algorithms, with a focus on extensions relevant to average-reward reinforcement learning. We first extend a stability proof method of Borkar and Meyn to accommodate more general noise conditions than previously considered, thereby yielding broader convergence guarantees for asynchronous SA. To sharpen the convergence analysis, we further examine the shadowing properties of asynchronous SA, building on a dynamical systems approach of Hirsch and Benaïm. These results provide a theoretical foundation for a class of relative value iteration-based reinforcement learning algorithms -- developed and analyzed in a companion paper -- for solving average-reward Markov and semi-Markov decision processes.

Xudong Lu, Zhi Zheng, Yi Wan et al.

Cross-View Geo-Localization (CVGL) focuses on identifying correspondences between images captured from distinct perspectives of the same geographical location. However, existing CVGL approaches are typically restricted to a single view or modality, and their direct visual matching strategy lacks interpretability: they only determine whether two images correspond, without explaining the rationale behind the match. In this paper, we present GLEAM-C, a foundational CVGL model that unifies multiple views and modalities-including UAV imagery, street maps, panoramic views, and ground photographs-by aligning them exclusively with satellite imagery. Our framework enhances training efficiency through optimized implementation while achieving accuracy comparable to prior modality-specific CVGL models through a two-phase training strategy. Moreover, to address the lack of interpretability in traditional CVGL methods, we leverage the reasoning capabilities of multimodal large language models (MLLMs) to propose a new task, GLEAM-X, which combines cross-view correspondence prediction with explainable reasoning. To support this task, we construct a bilingual benchmark using GPT-4o and Doubao-1.5-Thinking-Vision-Pro to generate training and testing data. The test set is further refined through detailed human revision, enabling systematic evaluation of explainable cross-view reasoning and advancing transparency and scalability in geo-localization. Together, GLEAM-C and GLEAM-X form a comprehensive CVGL pipeline that integrates multi-modal, multi-view alignment with interpretable correspondence analysis, unifying accurate cross-view matching with explainable reasoning and advancing Geo-Localization by enabling models to better Explain And Match. Code and datasets used in this work will be made publicly accessible at https://github.com/Lucky-Lance/GLEAM.

Yongjun Zhang, Mingtao Xiong, Yi Wan et al.

Cross-view localization (CVL) matches ground-level images with aerial references to determine the geo-position of a camera, enabling smart vehicles to self-localize offline in GNSS-denied environments. However, most CVL methods output only a single observation, the camera pose, and lack the redundant observations required by surveying principles, making it challenging to assess localization reliability through the mutual validation of observational data. To tackle this, we introduce Slice-Loc, a two-stage method featuring an a-contrario reliability validation for CVL. Instead of using the query image as a single input, Slice-Loc divides it into sub-images and estimates the 3-DoF pose for each slice, creating redundant and independent observations. Then, a geometric rigidity formula is proposed to filter out the erroneous 3-DoF poses, and the inliers are merged to generate the final camera pose. Furthermore, we propose a model that quantifies the meaningfulness of localization by estimating the number of false alarms (NFA), according to the distribution of the locations of the sliced images. By eliminating gross errors, Slice-Loc boosts localization accuracy and effectively detects failures. After filtering out mislocalizations, Slice-Loc reduces the proportion of errors exceeding 10 m to under 3\%. In cross-city tests on the DReSS dataset, Slice-Loc cuts the mean localization error from 4.47 m to 1.86 m and the mean orientation error from $\mathbf{3.42^{\circ}}$ to $\mathbf{1.24^{\circ}}$, outperforming state-of-the-art methods. Code and dataset will be available at: https://github.com/bnothing/Slice-Loc.

Youjie Zhou, Guofeng Mei, Yiming Wang et al.

Visual SLAM is particularly challenging in environments affected by noise, varying lighting conditions, and darkness. Learning-based optical flow algorithms can leverage multiple modalities to address these challenges, but traditional optical flow-based visual SLAM approaches often require significant computational resources.To overcome this limitation, we propose FMF-SLAM, an efficient multimodal fusion SLAM method that utilizes fast Fourier transform (FFT) to enhance the algorithm efficiency. Specifically, we introduce a novel Fourier-based self-attention and cross-attention mechanism to extract features from RGB and depth signals. We further enhance the interaction of multimodal features by incorporating multi-scale knowledge distillation across modalities. We also demonstrate the practical feasibility of FMF-SLAM in real-world scenarios with real time performance by integrating it with a security robot by fusing with a global positioning module GNSS-RTK and global Bundle Adjustment. Our approach is validated using video sequences from TUM, TartanAir, and our real-world datasets, showcasing state-of-the-art performance under noisy, varying lighting, and dark conditions.Our code and datasets are available at https://github.com/youjie-zhou/FMF-SLAM.git.

Yi Wan, Jiuqi Wang, Liam Li et al.

Tool-augmented large language models (LLMs) are emerging as deep research agents, systems that decompose complex queries, retrieve external evidence, and synthesize grounded responses. Yet current agents remain limited by shallow retrieval, weak alignment metrics, and brittle tool-use behavior. We introduce PokeeResearch-7B, a 7B-parameter deep research agent built under a unified reinforcement learning framework for robustness, alignment, and scalability. PokeeResearch-7B is trained by an annotation-free Reinforcement Learning from AI Feedback (RLAIF) framework to optimize policies using LLM-based reward signals that capture factual accuracy, citation faithfulness, and instruction adherence. A chain-of-thought-driven multi-call reasoning scaffold further enhances robustness through self-verification and adaptive recovery from tool failures. Among 10 popular deep research benchmarks, PokeeResearch-7B achieves state-of-the-art performance among 7B-scale deep research agents. This highlights that careful reinforcement learning and reasoning design can produce efficient, resilient, and research-grade AI agents. The model and inference code is open-sourced under Apache 2.0 license at https://github.com/Pokee-AI/PokeeResearchOSS.

Peiqi Chen, Lei Yu, Yi Wan et al.

Semi-dense feature matching methods have shown strong performance in challenging scenarios. However, the existing pipeline relies on a global search across the entire feature map to establish coarse matches, limiting further improvements in accuracy and efficiency. Motivated by this limitation, we propose a novel pipeline, CasP, which leverages cascaded correspondence priors for guidance. Specifically, the matching stage is decomposed into two progressive phases, bridged by a region-based selective cross-attention mechanism designed to enhance feature discriminability. In the second phase, one-to-one matches are determined by restricting the search range to the one-to-many prior areas identified in the first phase. Additionally, this pipeline benefits from incorporating high-level features, which helps reduce the computational costs of low-level feature extraction. The acceleration gains of CasP increase with higher resolution, and our lite model achieves a speedup of $\sim2.2\times$ at a resolution of 1152 compared to the most efficient method, ELoFTR. Furthermore, extensive experiments demonstrate its superiority in geometric estimation, particularly with impressive cross-domain generalization. These advantages highlight its potential for latency-sensitive and high-robustness applications, such as SLAM and UAV systems. Code is available at https://github.com/pq-chen/CasP.

Youjie Zhou, Yiming Wang, Fabio Poiesi et al.

We present a simple yet effective method to address loop closure detection in simultaneous localisation and mapping using local 3D deep descriptors (L3Ds). L3Ds are emerging compact representations of patches extracted from point clouds that are learnt from data using a deep learning algorithm. We propose a novel overlap measure for loop detection by computing the metric error between points that correspond to mutually-nearest-neighbour descriptors after registering the loop candidate point cloud by its estimated relative pose. This novel approach enables us to accurately detect loops and estimate six degrees-of-freedom poses in the case of small overlaps. We compare our L3D-based loop closure approach with recent approaches on LiDAR data and achieve state-of-the-art loop closure detection accuracy. Additionally, we embed our loop closure approach in RESLAM, a recent edge-based SLAM system, and perform the evaluation on real-world RGBD-TUM and synthetic ICL datasets. Our approach enables RESLAM to achieve a better localisation accuracy compared to its original loop closure strategy. Our project page is available at github.com/yiming107/l3d_loop_closure.

Wei Xu, Yi Wan

The attention mechanism has gained significant recognition in the field of computer vision due to its ability to effectively enhance the performance of deep neural networks. However, existing methods often struggle to effectively utilize spatial information or, if they do, they come at the cost of reducing channel dimensions or increasing the complexity of neural networks. In order to address these limitations, this paper introduces an Efficient Local Attention (ELA) method that achieves substantial performance improvements with a simple structure. By analyzing the limitations of the Coordinate Attention method, we identify the lack of generalization ability in Batch Normalization, the adverse effects of dimension reduction on channel attention, and the complexity of attention generation process. To overcome these challenges, we propose the incorporation of 1D convolution and Group Normalization feature enhancement techniques. This approach enables accurate localization of regions of interest by efficiently encoding two 1D positional feature maps without the need for dimension reduction, while allowing for a lightweight implementation. We carefully design three hyperparameters in ELA, resulting in four different versions: ELA-T, ELA-B, ELA-S, and ELA-L, to cater to the specific requirements of different visual tasks such as image classification, object detection and sementic segmentation. ELA can be seamlessly integrated into deep CNN networks such as ResNet, MobileNet, and DeepLab. Extensive evaluations on the ImageNet, MSCOCO, and Pascal VOC datasets demonstrate the superiority of the proposed ELA module over current state-of-the-art methods in all three aforementioned visual tasks.

Baochen Fu, Yuntao Du, Cheng Chang et al.

As real-world knowledge continues to evolve, the parametric knowledge acquired by multimodal models during pretraining becomes increasingly difficult to remain consistent with real-world knowledge. Existing research on multimodal knowledge updating focuses only on learning previously unknown knowledge, while overlooking the need to update knowledge that the model has already mastered but that later changes; moreover, evaluation is limited to the same modality, lacking a systematic analysis of cross-modal consistency. To address these issues, this paper proposes MMKU-Bench, a comprehensive evaluation benchmark for multimodal knowledge updating, which contains over 25k knowledge instances and more than 49k images, covering two scenarios, updated knowledge and unknown knowledge, thereby enabling comparative analysis of learning across different knowledge types. On this benchmark, we evaluate a variety of representative approaches, including supervised fine-tuning (SFT), reinforcement learning from human feedback (RLHF), and knowledge editing (KE). Experimental results show that SFT and RLHF are prone to catastrophic forgetting, while KE better preserve general capabilities but exhibit clear limitations in continual updating. Overall, MMKU-Bench provides a reliable and comprehensive evaluation benchmark for multimodal knowledge updating, advancing progress in this field.

Xiang He, Weiye Song, Yiming Wang et al.

Automatic retinal layer segmentation with medical images, such as optical coherence tomography (OCT) images, serves as an important tool for diagnosing ophthalmic diseases. However, it is challenging to achieve accurate segmentation due to low contrast and blood flow noises presented in the images. In addition, the algorithm should be light-weight to be deployed for practical clinical applications. Therefore, it is desired to design a light-weight network with high performance for retinal layer segmentation. In this paper, we propose LightReSeg for retinal layer segmentation which can be applied to OCT images. Specifically, our approach follows an encoder-decoder structure, where the encoder part employs multi-scale feature extraction and a Transformer block for fully exploiting the semantic information of feature maps at all scales and making the features have better global reasoning capabilities, while the decoder part, we design a multi-scale asymmetric attention (MAA) module for preserving the semantic information at each encoder scale. The experiments show that our approach achieves a better segmentation performance compared to the current state-of-the-art method TransUnet with 105.7M parameters on both our collected dataset and two other public datasets, with only 3.3M parameters.

Abhishek Naik, Yi Wan, Manan Tomar et al.

We show that discounted methods for solving continuing reinforcement learning problems can perform significantly better if they center their rewards by subtracting out the rewards' empirical average. The improvement is substantial at commonly used discount factors and increases further as the discount factor approaches one. In addition, we show that if a problem's rewards are shifted by a constant, then standard methods perform much worse, whereas methods with reward centering are unaffected. Estimating the average reward is straightforward in the on-policy setting; we propose a slightly more sophisticated method for the off-policy setting. Reward centering is a general idea, so we expect almost every reinforcement-learning algorithm to benefit by the addition of reward centering.

Qiong Wu, Panwang Xia, Lei Yu et al.

Cross-view geo-localization (CVGL) has been widely applied in fields such as robotic navigation and augmented reality. Existing approaches primarily use single images or fixed-view image sequences as queries, which limits perspective diversity. In contrast, when humans determine their location visually, they typically move around to gather multiple perspectives. This behavior suggests that integrating diverse visual cues can improve geo-localization reliability. Therefore, we propose a novel task: Cross-View Image Set Geo-Localization (Set-CVGL), which gathers multiple images with diverse perspectives as a query set for localization. To support this task, we introduce SetVL-480K, a benchmark comprising 480,000 ground images captured worldwide and their corresponding satellite images, with each satellite image corresponds to an average of 40 ground images from varied perspectives and locations. Furthermore, we propose FlexGeo, a flexible method designed for Set-CVGL that can also adapt to single-image and image-sequence inputs. FlexGeo includes two key modules: the Similarity-guided Feature Fuser (SFF), which adaptively fuses image features without prior content dependency, and the Individual-level Attributes Learner (IAL), leveraging geo-attributes of each image for comprehensive scene perception. FlexGeo consistently outperforms existing methods on SetVL-480K and two public datasets, SeqGeo and KITTI-CVL, achieving a localization accuracy improvement of over 22% on SetVL-480K.

Xuejun Huang, Xinyi Liu, Yi Wan et al.

Three-dimensional scene reconstruction from sparse-view satellite images is a long-standing and challenging task. While 3D Gaussian Splatting (3DGS) and its variants have recently attracted attention for its high efficiency, existing methods remain unsuitable for satellite images due to incompatibility with rational polynomial coefficient (RPC) models and limited generalization capability. Recent advances in generalizable 3DGS approaches show potential, but they perform poorly on multi-temporal sparse satellite images due to limited geometric constraints, transient objects, and radiometric inconsistencies. To address these limitations, we propose SkySplat, a novel self-supervised framework that integrates the RPC model into the generalizable 3DGS pipeline, enabling more effective use of sparse geometric cues for improved reconstruction. SkySplat relies only on RGB images and radiometric-robust relative height supervision, thereby eliminating the need for ground-truth height maps. Key components include a Cross-Self Consistency Module (CSCM), which mitigates transient object interference via consistency-based masking, and a multi-view consistency aggregation strategy that refines reconstruction results. Compared to per-scene optimization methods, SkySplat achieves an 86 times speedup over EOGS with higher accuracy. It also outperforms generalizable 3DGS baselines, reducing MAE from 13.18 m to 1.80 m on the DFC19 dataset significantly, and demonstrates strong cross-dataset generalization on the MVS3D benchmark.

Yi Liu, Xinyi Liu, Yi Wan et al.

Stereo image super-resolution (SSR) aims to enhance high-resolution details by leveraging information from stereo image pairs. However, existing stereo super-resolution (SSR) upsampling methods (e.g., pixel shuffle) often overlook cross-view geometric consistency and are limited to fixed-scale upsampling. The key issue is that previous upsampling methods use convolution to independently process deep features of different views, lacking cross-view and non-local information perception, making it difficult to select beneficial information from multi-view scenes adaptively. In this work, we propose Stereo Implicit Neural Representation (StereoINR), which innovatively models stereo image pairs as continuous implicit representations. This continuous representation breaks through the scale limitations, providing a unified solution for arbitrary-scale stereo super-resolution reconstruction of left-right views. Furthermore, by incorporating spatial warping and cross-attention mechanisms, StereoINR enables effective cross-view information fusion and achieves significant improvements in pixel-level geometric consistency. Extensive experiments across multiple datasets show that StereoINR outperforms out-of-training-distribution scale upsampling and matches state-of-the-art SSR methods within training-distribution scales.

Panwang Xia, Lei Yu, Yi Wan et al.

Cross-View Geo-Localization tackles the challenge of image geo-localization in GNSS-denied environments, including disaster response scenarios, urban canyons, and dense forests, by matching street-view query images with geo-tagged aerial-view reference images. However, current research often relies on benchmarks and methods that assume center-aligned settings or account for only limited decentrality, which we define as the offset of the query image relative to the reference image center. Such assumptions fail to reflect real-world scenarios, where reference databases are typically pre-established without the possibility of ensuring perfect alignment for each query image. Moreover, decentrality is a critical factor warranting deeper investigation, as larger decentrality can substantially improve localization efficiency but comes at the cost of declines in localization accuracy. To address this limitation, we introduce DReSS (Decentrality Related Street-view and Satellite-view dataset), a novel dataset designed to evaluate cross-view geo-localization with a large geographic scope and diverse landscapes, emphasizing the decentrality issue. Meanwhile, we propose AuxGeo (Auxiliary Enhanced Geo-Localization) to further study the decentrality issue, which leverages a multi-metric optimization strategy with two novel modules: the Bird's-eye view Intermediary Module (BIM) and the Position Constraint Module (PCM). These modules improve the localization accuracy despite the decentrality problem. Extensive experiments demonstrate that AuxGeo outperforms previous methods on our proposed DReSS dataset, mitigating the issue of large decentrality, and also achieves state-of-the-art performance on existing public datasets such as CVUSA, CVACT, and VIGOR.

Huizhen Yu, Yi Wan, Richard S. Sutton

This paper applies the authors' recent results on asynchronous stochastic approximation (SA) in the Borkar-Meyn framework to reinforcement learning in average-reward semi-Markov decision processes (SMDPs). We establish the convergence of an asynchronous SA analogue of Schweitzer's classical relative value iteration algorithm, RVI Q-learning, for finite-space, weakly communicating SMDPs. In particular, we show that the algorithm converges almost surely to a compact, connected subset of solutions to the average-reward optimality equation, with convergence to a unique, sample path-dependent solution under additional stepsize and asynchrony conditions. Moreover, to make full use of the SA framework, we introduce new monotonicity conditions for estimating the optimal reward rate in RVI Q-learning. These conditions substantially expand the previously considered algorithmic framework and are addressed through novel arguments in the stability and convergence analysis of RVI Q-learning.

Panwang Xia, Qiong Wu, Lei Yu et al.

Cross-view localization aims to estimate the 3-DoF pose of a ground-view image by aligning it with aerial or satellite imagery. Existing methods typically address this task through direct regression or feature alignment in a shared bird's-eye view (BEV) space. Although effective for coarse alignment, these methods fail to establish fine-grained and geometrically reliable correspondences under large viewpoint variations, thereby limiting both the accuracy and interpretability of localization results. Consequently, we revisit cross-view localization from the perspective of image matching and propose a unified framework that enhances both matching and localization. Specifically, we introduce a Surface Model that constrains BEV feature projection to physically valid regions for geometric consistency, and a SimRefiner that adaptively refines similarity distributions to enhance match reliability. To further support research in this area, we present CVFM, the first benchmark with 32,509 cross-view image pairs annotated with pixel-level correspondences. Extensive experiments demonstrate that our approach achieves geometry-consistent and fine-grained correspondences across extreme viewpoints and further improves the accuracy and stability of cross-view localization.

Yi Wan, Dmytro Korenkevych, Zheqing Zhu

In reinforcement learning (RL), continuing tasks refer to tasks where the agent-environment interaction is ongoing and can not be broken down into episodes. These tasks are suitable when environment resets are unavailable, agent-controlled, or predefined but where all rewards-including those beyond resets-are critical. These scenarios frequently occur in real-world applications and can not be modeled by episodic tasks. While modern deep RL algorithms have been extensively studied and well understood in episodic tasks, their behavior in continuing tasks remains underexplored. To address this gap, we provide an empirical study of several well-known deep RL algorithms using a suite of continuing task testbeds based on Mujoco and Atari environments, highlighting several key insights concerning continuing tasks. Using these testbeds, we also investigate the effectiveness of a method for improving temporal-difference-based RL algorithms in continuing tasks by centering rewards, as introduced by Naik et al. (2024). While their work primarily focused on this method in conjunction with Q-learning, our results extend their findings by demonstrating that this method is effective across a broader range of algorithms, scales to larger tasks, and outperforms two other reward-centering approaches.

Huizhen Yu, Yi Wan, Richard S. Sutton

In this paper, we study asynchronous stochastic approximation algorithms without communication delays. Our main contribution is a stability proof for these algorithms that extends a method of Borkar and Meyn by accommodating more general noise conditions. We also derive convergence results from this stability result and discuss their application in important average-reward reinforcement learning problems.

Yongjun Zhang, Siyuan Zou, Xinyi Liu et al.

The complementary fusion of light detection and ranging (LiDAR) data and image data is a promising but challenging task for generating high-precision and high-density point clouds. This study proposes an innovative LiDAR-guided stereo matching approach called LiDAR-guided stereo matching (LGSM), which considers the spatial consistency represented by continuous disparity or depth changes in the homogeneous region of an image. The LGSM first detects the homogeneous pixels of each LiDAR projection point based on their color or intensity similarity. Next, we propose a riverbed enhancement function to optimize the cost volume of the LiDAR projection points and their homogeneous pixels to improve the matching robustness. Our formulation expands the constraint scopes of sparse LiDAR projection points with the guidance of image information to optimize the cost volume of pixels as much as possible. We applied LGSM to semi-global matching and AD-Census on both simulated and real datasets. When the percentage of LiDAR points in the simulated datasets was 0.16%, the matching accuracy of our method achieved a subpixel level, while that of the original stereo matching algorithm was 3.4 pixels. The experimental results show that LGSM is suitable for indoor, street, aerial, and satellite image datasets and provides good transferability across semi-global matching and AD-Census. Furthermore, the qualitative and quantitative evaluations demonstrate that LGSM is superior to two state-of-the-art optimizing cost volume methods, especially in reducing mismatches in difficult matching areas and refining the boundaries of objects.

Yi Wan, Abhishek Naik, Richard S. Sutton

We extend the options framework for temporal abstraction in reinforcement learning from discounted Markov decision processes (MDPs) to average-reward MDPs. Our contributions include general convergent off-policy inter-option learning algorithms, intra-option algorithms for learning values and models, as well as sample-based planning variants of our learning algorithms. Our algorithms and convergence proofs extend those recently developed by Wan, Naik, and Sutton. We also extend the notion of option-interrupting behavior from the discounted to the average-reward formulation. We show the efficacy of the proposed algorithms with experiments on a continuing version of the Four-Room domain.

Katya Kudashkina, Yi Wan, Abhishek Naik et al.

In model-based reinforcement learning (MBRL), Wan et al. (2019) showed conditions under which the environment model could produce the expectation of the next feature vector rather than the full distribution, or a sample thereof, with no loss in planning performance. Such expectation models are of interest when the environment is stochastic and non-stationary, and the model is approximate, such as when it is learned using function approximation. In these cases a full distribution model may be impractical and a sample model may be either more expensive computationally or of high variance. Wan et al. considered only planning for prediction to evaluate a fixed policy. In this paper, we treat the control case - planning to improve and find a good approximate policy. We prove that planning with an expectation model must update a state-value function, not an action-value function as previously suggested (e.g., Sorg & Singh, 2010). This opens the question of how planning influences action selections. We consider three strategies for this and present general MBRL algorithms for each. We identify the strengths and weaknesses of these algorithms in computational experiments. Our algorithms and experiments are the first to treat MBRL with expectation models in a general setting.

Shangtong Zhang, Yi Wan, Richard S. Sutton et al.

We consider off-policy policy evaluation with function approximation (FA) in average-reward MDPs, where the goal is to estimate both the reward rate and the differential value function. For this problem, bootstrapping is necessary and, along with off-policy learning and FA, results in the deadly triad (Sutton & Barto, 2018). To address the deadly triad, we propose two novel algorithms, reproducing the celebrated success of Gradient TD algorithms in the average-reward setting. In terms of estimating the differential value function, the algorithms are the first convergent off-policy linear function approximation algorithms. In terms of estimating the reward rate, the algorithms are the first convergent off-policy linear function approximation algorithms that do not require estimating the density ratio. We demonstrate empirically the advantage of the proposed algorithms, as well as their nonlinear variants, over a competitive density-ratio-based approach, in a simple domain as well as challenging robot simulation tasks.

Yicheng Guo, Yujin Wen, Congwei Jiang et al.

Anomaly detection is a crucial and challenging subject that has been studied within diverse research areas. In this work, we explore the task of log anomaly detection (especially computer system logs and user behavior logs) by analyzing logs' sequential information. We propose LAMA, a multi-head attention based sequential model to process log streams as template activity (event) sequences. A next event prediction task is applied to train the model for anomaly detection. Extensive empirical studies demonstrate that our new model outperforms existing log anomaly detection methods including statistical and deep learning methodologies, which validate the effectiveness of our proposed method in learning sequence patterns of log data.

Yi Wan, Abhishek Naik, Richard S. Sutton

We introduce learning and planning algorithms for average-reward MDPs, including 1) the first general proven-convergent off-policy model-free control algorithm without reference states, 2) the first proven-convergent off-policy model-free prediction algorithm, and 3) the first off-policy learning algorithm that converges to the actual value function rather than to the value function plus an offset. All of our algorithms are based on using the temporal-difference error rather than the conventional error when updating the estimate of the average reward. Our proof techniques are a slight generalization of those by Abounadi, Bertsekas, and Borkar (2001). In experiments with an Access-Control Queuing Task, we show some of the difficulties that can arise when using methods that rely on reference states and argue that our new algorithms can be significantly easier to use.

Zhimin Hou, Kuangen Zhang, Yi Wan et al.

The optimal policy of a reinforcement learning problem is often discontinuous and non-smooth. I.e., for two states with similar representations, their optimal policies can be significantly different. In this case, representing the entire policy with a function approximator (FA) with shared parameters for all states maybe not desirable, as the generalization ability of parameters sharing makes representing discontinuous, non-smooth policies difficult. A common way to solve this problem, known as Mixture-of-Experts, is to represent the policy as the weighted sum of multiple components, where different components perform well on different parts of the state space. Following this idea and inspired by a recent work called advantage-weighted information maximization, we propose to learn for each state weights of these components, so that they entail the information of the state itself and also the preferred action learned so far for the state. The action preference is characterized via the advantage function. In this case, the weight of each component would only be large for certain groups of states whose representations are similar and preferred action representations are also similar. Therefore each component is easy to be represented. We call a policy parameterized in this way an Advantage Weighted Mixture Policy (AWMP) and apply this idea to improve soft-actor-critic (SAC), one of the most competitive continuous control algorithm. Experimental results demonstrate that SAC with AWMP clearly outperforms SAC in four commonly used continuous control tasks and achieve stable performance across different random seeds.

Yi Wan, Zaheer Abbas, Adam White et al.

Distribution and sample models are two popular model choices in model-based reinforcement learning (MBRL). However, learning these models can be intractable, particularly when the state and action spaces are large. Expectation models, on the other hand, are relatively easier to learn due to their compactness and have also been widely used for deterministic environments. For stochastic environments, it is not obvious how expectation models can be used for planning as they only partially characterize a distribution. In this paper, we propose a sound way of using approximate expectation models for MBRL. In particular, we 1) show that planning with an expectation model is equivalent to planning with a distribution model if the state value function is linear in state features, 2) analyze two common parametrization choices for approximating the expectation: linear and non-linear expectation models, 3) propose a sound model-based policy evaluation algorithm and present its convergence results, and 4) empirically demonstrate the effectiveness of the proposed planning algorithm.

Hao Wu, Yi Wan

In computer vision, the estimation of the fundamental matrix is a basic problem that has been extensively studied. The accuracy of the estimation imposes a significant influence on subsequent tasks such as the camera trajectory determination and 3D reconstruction. In this paper we propose a new method for fundamental matrix estimation that makes use of clustering a group of 4D vectors. The key insight is the observation that among the 4D vectors constructed from matching pairs of points obtained from the SIFT algorithm, well-defined cluster points tend to be reliable inliers suitable for fundamental matrix estimation. Based on this, we utilizes a recently proposed efficient clustering method through density peaks seeking and propose a new clustering assisted method. Experimental results show that the proposed algorithm is faster and more accurate than currently commonly used methods.