The BrowserGym Ecosystem for Web Agent Research

The BrowserGym ecosystem addresses the growing need for efficient evaluation and benchmarking of web agents, particularly those leveraging automation and Large Language Models (LLMs) for web interaction tasks. Many existing benchmarks suffer from fragmentation and inconsistent evaluation methodologies, making it challenging to achieve reliable comparisons and reproducible results. BrowserGym aims to solve this by providing a unified, gym-like environment with well-defined observation and action spaces, facilitating standardized evaluation across diverse benchmarks. Combined with AgentLab, a complementary framework that aids in agent creation, testing, and analysis, BrowserGym offers flexibility for integrating new benchmarks while ensuring consistent evaluation and comprehensive experiment management. This standardized approach seeks to reduce the time and complexity of developing web agents, supporting more reliable comparisons and facilitating in-depth analysis of agent behaviors, and could result in more adaptable, capable agents, ultimately accelerating innovation in LLM-driven automation. As a supporting evidence, we conduct the first large-scale, multi-benchmark web agent experiment and compare the performance of 6 state-of-the-art LLMs across all benchmarks currently available in BrowserGym. Among other findings, our results highlight a large discrepancy between OpenAI and Anthropic's latests models, with Claude-3.5-Sonnet leading the way on almost all benchmarks, except on vision-related tasks where GPT-4o is superior. Despite these advancements, our results emphasize that building robust and efficient web agents remains a significant challenge, due to the inherent complexity of real-world web environments and the limitations of current models.

Blind Inverse Problem Solving Made Easy by Text-to-Image Latent Diffusion

Blind inverse problems, where both the target data and forward operator are unknown, are crucial to many computer vision applications. Existing methods often depend on restrictive assumptions such as additional training, operator linearity, or narrow image distributions, thus limiting their generalizability. In this work, we present LADiBI, a training-free framework that uses large-scale text-to-image diffusion models to solve blind inverse problems with minimal assumptions. By leveraging natural language prompts, LADiBI jointly models priors for both the target image and operator, allowing for flexible adaptation across a variety of tasks. Additionally, we propose a novel posterior sampling approach that combines effective operator initialization with iterative refinement, enabling LADiBI to operate without predefined operator forms. Our experiments show that LADiBI is capable of solving a broad range of image restoration tasks, including both linear and nonlinear problems, on diverse target image distributions.

Local Policies Enable Zero-shot Long-horizon Manipulation

Sim2real for robotic manipulation is difficult due to the challenges of simulating complex contacts and generating realistic task distributions. To tackle the latter problem, we introduce ManipGen, which leverages a new class of policies for sim2real transfer: local policies. Locality enables a variety of appealing properties including invariances to absolute robot and object pose, skill ordering, and global scene configuration. We combine these policies with foundation models for vision, language and motion planning and demonstrate SOTA zero-shot performance of our method to Robosuite benchmark tasks in simulation (97%). We transfer our local policies from simulation to reality and observe they can solve unseen long-horizon manipulation tasks with up to 8 stages with significant pose, object and scene configuration variation. ManipGen outperforms SOTA approaches such as SayCan, OpenVLA, LLMTrajGen and VoxPoser across 50 real-world manipulation tasks by 36%, 76%, 62% and 60% respectively. Video results at https://mihdalal.github.io/manipgen/

Embodied-RAG: General non-parametric Embodied Memory for Retrieval and Generation

There is no limit to how much a robot might explore and learn, but all of that knowledge needs to be searchable and actionable. Within language research, retrieval augmented generation (RAG) has become the workhouse of large-scale non-parametric knowledge, however existing techniques do not directly transfer to the embodied domain, which is multimodal, data is highly correlated, and perception requires abstraction. To address these challenges, we introduce Embodied-RAG, a framework that enhances the foundational model of an embodied agent with a non-parametric memory system capable of autonomously constructing hierarchical knowledge for both navigation and language generation. Embodied-RAG handles a full range of spatial and semantic resolutions across diverse environments and query types, whether for a specific object or a holistic description of ambiance. At its core, Embodied-RAG's memory is structured as a semantic forest, storing language descriptions at varying levels of detail. This hierarchical organization allows the system to efficiently generate context-sensitive outputs across different robotic platforms. We demonstrate that Embodied-RAG effectively bridges RAG to the robotics domain, successfully handling over 200 explanation and navigation queries across 19 environments, highlighting its promise for general-purpose non-parametric system for embodied agents.

Neural MP: A Generalist Neural Motion Planner

The current paradigm for motion planning generates solutions from scratch for every new problem, which consumes significant amounts of time and computational resources. For complex, cluttered scenes, motion planning approaches can often take minutes to produce a solution, while humans are able to accurately and safely reach any goal in seconds by leveraging their prior experience. We seek to do the same by applying data-driven learning at scale to the problem of motion planning. Our approach builds a large number of complex scenes in simulation, collects expert data from a motion planner, then distills it into a reactive generalist policy. We then combine this with lightweight optimization to obtain a safe path for real world deployment. We perform a thorough evaluation of our method on 64 motion planning tasks across four diverse environments with randomized poses, scenes and obstacles, in the real world, demonstrating an improvement of 23%, 17% and 79% motion planning success rate over state of the art sampling, optimization and learning based planning methods. Video results available at mihdalal.github.io/neuralmotionplanner

Situated Instruction Following

Language is never spoken in a vacuum. It is expressed, comprehended, and contextualized within the holistic backdrop of the speaker's history, actions, and environment. Since humans are used to communicating efficiently with situated language, the practicality of robotic assistants hinge on their ability to understand and act upon implicit and situated instructions. In traditional instruction following paradigms, the agent acts alone in an empty house, leading to language use that is both simplified and artificially "complete." In contrast, we propose situated instruction following, which embraces the inherent underspecification and ambiguity of real-world communication with the physical presence of a human speaker. The meaning of situated instructions naturally unfold through the past actions and the expected future behaviors of the human involved. Specifically, within our settings we have instructions that (1) are ambiguously specified, (2) have temporally evolving intent, (3) can be interpreted more precisely with the agent's dynamic actions. Our experiments indicate that state-of-the-art Embodied Instruction Following (EIF) models lack holistic understanding of situated human intention.

HEMM: Holistic Evaluation of Multimodal Foundation Models

Multimodal foundation models that can holistically process text alongside images, video, audio, and other sensory modalities are increasingly used in a variety of real-world applications. However, it is challenging to characterize and study progress in multimodal foundation models, given the range of possible modeling decisions, tasks, and domains. In this paper, we introduce Holistic Evaluation of Multimodal Models (HEMM) to systematically evaluate the capabilities of multimodal foundation models across a set of 3 dimensions: basic skills, information flow, and real-world use cases. Basic multimodal skills are internal abilities required to solve problems, such as learning interactions across modalities, fine-grained alignment, multi-step reasoning, and the ability to handle external knowledge. Information flow studies how multimodal content changes during a task through querying, translation, editing, and fusion. Use cases span domain-specific challenges introduced in real-world multimedia, affective computing, natural sciences, healthcare, and human-computer interaction applications. Through comprehensive experiments across the 30 tasks in HEMM, we (1) identify key dataset dimensions (e.g., basic skills, information flows, and use cases) that pose challenges to today's models, and (2) distill performance trends regarding how different modeling dimensions (e.g., scale, pre-training data, multimodal alignment, pre-training, and instruction tuning objectives) influence performance. Our conclusions regarding challenging multimodal interactions, use cases, and tasks requiring reasoning and external knowledge, the benefits of data and model scale, and the impacts of instruction tuning yield actionable insights for future work in multimodal foundation models.

Tree Search for Language Model Agents

Autonomous agents powered by language models (LMs) have demonstrated promise in their ability to perform decision-making tasks such as web automation. However, a key limitation remains: LMs, primarily optimized for natural language understanding and generation, struggle with multi-step reasoning, planning, and using environmental feedback when attempting to solve realistic computer tasks. Towards addressing this, we propose an inference-time search algorithm for LM agents to explicitly perform exploration and multi-step planning in interactive web environments. Our approach is a form of best-first tree search that operates within the actual environment space, and is complementary with most existing state-of-the-art agents. It is the first tree search algorithm for LM agents that shows effectiveness on realistic web tasks. On the challenging VisualWebArena benchmark, applying our search algorithm on top of a GPT-4o agent yields a 39.7% relative increase in success rate compared to the same baseline without search, setting a state-of-the-art success rate of 26.4%. On WebArena, search also yields a 28.0% relative improvement over a baseline agent, setting a competitive success rate of 19.2%. Our experiments highlight the effectiveness of search for web agents, and we demonstrate that performance scales with increased test-time compute. We conduct a thorough analysis of our results to highlight improvements from search, limitations, and promising directions for future work. Our code and models are publicly released at https://jykoh.com/search-agents.

Adversarial Attacks on Multimodal Agents

Vision-enabled language models (VLMs) are now used to build autonomous multimodal agents capable of taking actions in real environments. In this paper, we show that multimodal agents raise new safety risks, even though attacking agents is more challenging than prior attacks due to limited access to and knowledge about the environment. Our attacks use adversarial text strings to guide gradient-based perturbation over one trigger image in the environment: (1) our captioner attack attacks white-box captioners if they are used to process images into captions as additional inputs to the VLM; (2) our CLIP attack attacks a set of CLIP models jointly, which can transfer to proprietary VLMs. To evaluate the attacks, we curated VisualWebArena-Adv, a set of adversarial tasks based on VisualWebArena, an environment for web-based multimodal agent tasks. Within an L-infinity norm of $16/256$ on a single image, the captioner attack can make a captioner-augmented GPT-4V agent execute the adversarial goals with a 75% success rate. When we remove the captioner or use GPT-4V to generate its own captions, the CLIP attack can achieve success rates of 21% and 43%, respectively. Experiments on agents based on other VLMs, such as Gemini-1.5, Claude-3, and GPT-4o, show interesting differences in their robustness. Further analysis reveals several key factors contributing to the attack's success, and we also discuss the implications for defenses as well. Project page: https://chenwu.io/attack-agent Code and data: https://github.com/ChenWu98/agent-attack

Understanding Visual Concepts Across Models

Large multimodal models such as Stable Diffusion can generate, detect, and classify new visual concepts after fine-tuning just a single word embedding. Do models learn similar words for the same concepts (i.e. <orange-cat> = orange + cat)? We conduct a large-scale analysis on three state-of-the-art models in text-to-image generation, open-set object detection, and zero-shot classification, and find that new word embeddings are model-specific and non-transferable. Across 4,800 new embeddings trained for 40 diverse visual concepts on four standard datasets, we find perturbations within an $\epsilon$-ball to any prior embedding that generate, detect, and classify an arbitrary concept. When these new embeddings are spliced into new models, fine-tuning that targets the original model is lost. We show popular soft prompt-tuning approaches find these perturbative solutions when applied to visual concept learning tasks, and embeddings for visual concepts are not transferable. Code for reproducing our work is available at: https://visual-words.github.io.