Explorer: Scaling Exploration-driven Web Trajectory Synthesis for Multimodal Web Agents

Recent success in large multimodal models (LMMs) has sparked promising applications of agents capable of autonomously completing complex web tasks. While open-source LMM agents have made significant advances in offline evaluation benchmarks, their performance still falls substantially short of human-level capabilities in more realistic online settings. A key bottleneck is the lack of diverse and large-scale trajectory-level datasets across various domains, which are expensive to collect. In this paper, we address this challenge by developing a scalable recipe to synthesize the largest and most diverse trajectory-level dataset to date, containing over 94K successful multimodal web trajectories, spanning 49K unique URLs, 720K screenshots, and 33M web elements. In particular, we leverage extensive web exploration and refinement to obtain diverse task intents. The average cost is 28 cents per successful trajectory, making it affordable to a wide range of users in the community. Leveraging this dataset, we train Explorer, a multimodal web agent, and demonstrate strong performance on both offline and online web agent benchmarks such as Mind2Web-Live, Multimodal-Mind2Web, and MiniWob++. Additionally, our experiments highlight data scaling as a key driver for improving web agent capabilities. We hope this study makes state-of-the-art LMM-based agent research at a larger scale more accessible.

Sparse Autoencoders for Scientifically Rigorous Interpretation of Vision Models

To truly understand vision models, we must not only interpret their learned features but also validate these interpretations through controlled experiments. Current approaches either provide interpretable features without the ability to test their causal influence, or enable model editing without interpretable controls. We present a unified framework using sparse autoencoders (SAEs) that bridges this gap, allowing us to discover human-interpretable visual features and precisely manipulate them to test hypotheses about model behavior. By applying our method to state-of-the-art vision models, we reveal key differences in the semantic abstractions learned by models with different pre-training objectives. We then demonstrate the practical usage of our framework through controlled interventions across multiple vision tasks. We show that SAEs can reliably identify and manipulate interpretable visual features without model re-training, providing a powerful tool for understanding and controlling vision model behavior. We provide code, demos and models on our project website: https://osu-nlp-group.github.io/SAE-V.

Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation

Class activation map (CAM) has been widely used to highlight image regions that contribute to class predictions. Despite its simplicity and computational efficiency, CAM often struggles to identify discriminative regions that distinguish visually similar fine-grained classes. Prior efforts address this limitation by introducing more sophisticated explanation processes, but at the cost of extra complexity. In this paper, we propose Finer-CAM, a method that retains CAM's efficiency while achieving precise localization of discriminative regions. Our key insight is that the deficiency of CAM lies not in "how" it explains, but in "what" it explains}. Specifically, previous methods attempt to identify all cues contributing to the target class's logit value, which inadvertently also activates regions predictive of visually similar classes. By explicitly comparing the target class with similar classes and spotting their differences, Finer-CAM suppresses features shared with other classes and emphasizes the unique, discriminative details of the target class. Finer-CAM is easy to implement, compatible with various CAM methods, and can be extended to multi-modal models for accurate localization of specific concepts. Additionally, Finer-CAM allows adjustable comparison strength, enabling users to selectively highlight coarse object contours or fine discriminative details. Quantitatively, we show that masking out the top 5% of activated pixels by Finer-CAM results in a larger relative confidence drop compared to baselines. The source code and demo are available at https://github.com/Imageomics/Finer-CAM.

Prompt-CAM: A Simpler Interpretable Transformer for Fine-Grained Analysis

We present a simple usage of pre-trained Vision Transformers (ViTs) for fine-grained analysis, aiming to identify and localize the traits that distinguish visually similar categories, such as different bird species or dog breeds. Pre-trained ViTs such as DINO have shown remarkable capabilities to extract localized, informative features. However, using saliency maps like Grad-CAM can hardly point out the traits: they often locate the whole object by a blurred, coarse heatmap, not traits. We propose a novel approach Prompt Class Attention Map (Prompt-CAM) to the rescue. Prompt-CAM learns class-specific prompts to a pre-trained ViT and uses the corresponding outputs for classification. To classify an image correctly, the true-class prompt must attend to the unique image patches not seen in other classes' images, i.e., traits. As such, the true class's multi-head attention maps reveal traits and their locations. Implementation-wise, Prompt-CAM is almost a free lunch by simply modifying the prediction head of Visual Prompt Tuning (VPT). This makes Prompt-CAM fairly easy to train and apply, sharply contrasting other interpretable methods that design specific models and training processes. It is even simpler than the recently published INterpretable TRansformer (INTR), whose encoder-decoder architecture prevents it from leveraging pre-trained ViTs. Extensive empirical studies on a dozen datasets from various domains (e.g., birds, fishes, insects, fungi, flowers, food, and cars) validate Prompt-CAM superior interpretation capability.

Static Segmentation by Tracking: A Frustratingly Label-Efficient Approach to Fine-Grained Segmentation

We study image segmentation in the biological domain, particularly trait and part segmentation from specimen images (e.g., butterfly wing stripes or beetle body parts). This is a crucial, fine-grained task that aids in understanding the biology of organisms. The conventional approach involves hand-labeling masks, often for hundreds of images per species, and training a segmentation model to generalize these labels to other images, which can be exceedingly laborious. We present a label-efficient method named Static Segmentation by Tracking (SST). SST is built upon the insight: while specimens of the same species have inherent variations, the traits and parts we aim to segment show up consistently. This motivates us to concatenate specimen images into a ``pseudo-video'' and reframe trait and part segmentation as a tracking problem. Concretely, SST generates masks for unlabeled images by propagating annotated or predicted masks from the ``pseudo-preceding'' images. Powered by Segment Anything Model 2 (SAM~2) initially developed for video segmentation, we show that SST can achieve high-quality trait and part segmentation with merely one labeled image per species -- a breakthrough for analyzing specimen images. We further develop a cycle-consistent loss to fine-tune the model, again using one labeled image. Additionally, we highlight the broader potential of SST, including one-shot instance segmentation on images taken in the wild and trait-based image retrieval.

RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics

Spatial understanding is a crucial capability for robots to make grounded decisions based on their environment. This foundational skill enables robots not only to perceive their surroundings but also to reason about and interact meaningfully within the world. In modern robotics, these capabilities are taken on by visual language models, and they face significant challenges when applied to spatial reasoning context due to their training data sources. These sources utilize general-purpose image datasets, and they often lack sophisticated spatial scene understanding capabilities. For example, the datasets do not address reference frame comprehension - spatial relationships require clear contextual understanding, whether from an ego-centric, object-centric, or world-centric perspective, which allow for effective real-world interaction. To address this issue, we introduce RoboSpatial, a large-scale spatial understanding dataset consisting of real indoor and tabletop scenes captured as 3D scans and egocentric images, annotated with rich spatial information relevant to robotics. The dataset includes 1M images, 5K 3D scans, and 3M annotated spatial relationships, with paired 2D egocentric images and 3D scans to make it both 2D and 3D ready. Our experiments show that models trained with RoboSpatial outperform baselines on downstream tasks such as spatial affordance prediction, spatial relationship prediction, and robotics manipulation.

Is Your LLM Secretly a World Model of the Internet? Model-Based Planning for Web Agents

Language agents have demonstrated promising capabilities in automating web-based tasks, though their current reactive approaches still underperform largely compared to humans. While incorporating advanced planning algorithms, particularly tree search methods, could enhance these agents' performance, implementing tree search directly on live websites poses significant safety risks and practical constraints due to irreversible actions such as confirming a purchase. In this paper, we introduce a novel paradigm that augments language agents with model-based planning, pioneering the innovative use of large language models (LLMs) as world models in complex web environments. Our method, WebDreamer, builds on the key insight that LLMs inherently encode comprehensive knowledge about website structures and functionalities. Specifically, WebDreamer uses LLMs to simulate outcomes for each candidate action (e.g., "what would happen if I click this button?") using natural language descriptions, and then evaluates these imagined outcomes to determine the optimal action at each step. Empirical results on two representative web agent benchmarks with online interaction -- VisualWebArena and Mind2Web-live -- demonstrate that WebDreamer achieves substantial improvements over reactive baselines. By establishing the viability of LLMs as world models in web environments, this work lays the groundwork for a paradigm shift in automated web interaction. More broadly, our findings open exciting new avenues for future research into 1) optimizing LLMs specifically for world modeling in complex, dynamic environments, and 2) model-based speculative planning for language agents.

Integrated Location Sensing and Communication for Ultra-Massive MIMO With Hybrid-Field Beam-Squint Effect

The advent of ultra-massive multiple-input-multiple output systems holds great promise for next-generation communications, yet their channels exhibit hybrid far- and near- field beam-squint (HFBS) effect. In this paper, we not only overcome but also harness the HFBS effect to propose an integrated location sensing and communication (ILSC) framework. During the uplink training stage, user terminals (UTs) transmit reference signals for simultaneous channel estimation and location sensing. This stage leverages an elaborately designed hybrid-field projection matrix to overcome the HFBS effect and estimate the channel in compressive manner. Subsequently, the scatterers' locations can be sensed from the spherical wavefront based on the channel estimation results. By treating the sensed scatterers as virtual anchors, we employ a weighted least-squares approach to derive UT' s location. Moreover, we propose an iterative refinement mechanism, which utilizes the accurately estimated time difference of arrival of multipath components to enhance location sensing precision. In the following downlink data transmission stage, we leverage the acquired location information to further optimize the hybrid beamformer, which combines the beam broadening and focusing to mitigate the spectral efficiency degradation resulted from the HFBS effect. Extensive simulation experiments demonstrate that the proposed ILSC scheme has superior location sensing and communication performance than conventional methods.

Navigating the Digital World as Humans Do: Universal Visual Grounding for GUI Agents

Multimodal large language models (MLLMs) are transforming the capabilities of graphical user interface (GUI) agents, facilitating their transition from controlled simulations to complex, real-world applications across various platforms. However, the effectiveness of these agents hinges on the robustness of their grounding capability. Current GUI agents predominantly utilize text-based representations such as HTML or accessibility trees, which, despite their utility, often introduce noise, incompleteness, and increased computational overhead. In this paper, we advocate a human-like embodiment for GUI agents that perceive the environment entirely visually and directly take pixel-level operations on the GUI. The key is visual grounding models that can accurately map diverse referring expressions of GUI elements to their coordinates on the GUI across different platforms. We show that a simple recipe, which includes web-based synthetic data and slight adaptation of the LLaVA architecture, is surprisingly effective for training such visual grounding models. We collect the largest dataset for GUI visual grounding so far, containing 10M GUI elements and their referring expressions over 1.3M screenshots, and use it to train UGround, a strong universal visual grounding model for GUI agents. Empirical results on six benchmarks spanning three categories (grounding, offline agent, and online agent) show that 1) UGround substantially outperforms existing visual grounding models for GUI agents, by up to 20% absolute, and 2) agents with UGround outperform state-of-the-art agents, despite the fact that existing agents use additional text-based input while ours only uses visual perception. These results provide strong support for the feasibility and promises of GUI agents that navigate the digital world as humans do.

ScienceAgentBench: Toward Rigorous Assessment of Language Agents for Data-Driven Scientific Discovery

The advancements of language language models (LLMs) have piqued growing interest in developing LLM-based language agents to automate scientific discovery end-to-end, which has sparked both excitement and skepticism about the true capabilities of such agents. In this work, we argue that for an agent to fully automate scientific discovery, it must be able to complete all essential tasks in the workflow. Thus, we call for rigorous assessment of agents on individual tasks in a scientific workflow before making bold claims on end-to-end automation. To this end, we present ScienceAgentBench, a new benchmark for evaluating language agents for data-driven scientific discovery. To ensure the scientific authenticity and real-world relevance of our benchmark, we extract 102 tasks from 44 peer-reviewed publications in four disciplines and engage nine subject matter experts to validate them. We unify the target output for every task to a self-contained Python program file and employ an array of evaluation metrics to examine the generated programs, execution results, and costs. Each task goes through multiple rounds of manual validation by annotators and subject matter experts to ensure its annotation quality and scientific plausibility. We also propose two effective strategies to mitigate data contamination concerns. Using our benchmark, we evaluate five open-weight and proprietary LLMs, each with three frameworks: direct prompting, OpenHands, and self-debug. Given three attempts for each task, the best-performing agent can only solve 32.4% of the tasks independently and 34.3% with expert-provided knowledge. These results underscore the limited capacities of current language agents in generating code for data-driven discovery, let alone end-to-end automation for scientific research.