Farama-Foundation/MiniWoB-plusplus

A collection of reinforcement learning environments for simple web interaction tasks

What it solves

MiniWoB++ provides a standardized way to train and test AI agents to interact with web browsers. It solves the difficulty of creating diverse, controlled environments where an agent can learn to perform specific web-based tasks (like clicking buttons or filling forms) without needing to build a full-scale website from scratch for every experiment.

How it works

The library offers over 100 web interaction environments. It uses the Gymnasium API for reinforcement learning and leverages Selenium WebDriver to programmatically control a Chrome or Chromium browser. Agents receive observations (such as the DOM elements and text instructions) and send back actions to interact with the page.

Who it’s for

Researchers and developers working on reinforcement learning, web automation, and AI agents capable of navigating and interacting with digital interfaces.

Highlights

  • Over 100 diverse web interaction environments.
  • Compatible with the Gymnasium API for easy integration with RL algorithms.
  • Uses Selenium WebDriver for real-world browser interaction.
  • Provides both JavaScript and Python interfaces.

Related

  • Project
    ServiceNow/BrowserGym

    An open-source framework that provides a unified interface for researching and evaluating web agents across multiple benchmarks and open-ended web tasks.

  • Project
    browser-use/web-ui

    A Gradio-based web interface for the browser-use library that enables users to interact with AI browser agents using a variety of LLMs.

  • Project
    Farama-Foundation/HighwayEnv

    A collection of Gymnasium-compatible environments for autonomous driving and tactical decision-making tasks.

  • Project
    allenai/ScienceWorld

    A text-based virtual environment for benchmarking AI agents on tasks derived from a standardized elementary science curriculum.

  • Project
    google-research/android_world

    An environment and benchmark for building and testing autonomous agents that can control Android devices across a wide variety of real-world apps.