Automated Reinforcement Learning · on the web
AutoRL X Labs
An open-source platform for Automated Reinforcement Learning. Build gyms, run optimizations, and visually inspect agent behavior — all directly in your browser.
The basics
What is Reinforcement Learning?
Reinforcement learning (RL) is a branch of machine learning for solving sequential decision-making problems. An agent interacts with an environment, takes actions, and learns from the rewards it receives — refining its strategy through trial and error until it finds behavior that maximizes long-term reward. The same idea powers advances in autonomous driving, robotics, healthcare, finance, and gaming.
Agent
The learner and decision-maker that observes the world and chooses what to do next.
Environment
The world the agent acts in. It reacts to each action and returns a new state.
Action
A move the agent can make. Its policy maps each state to the action to take.
Reward
Feedback on how good an action was. The agent learns to maximize it over time.
AutoRL X takes this loop further with Automated RL (AutoRL) — helping automate algorithm selection, hyperparameter tuning, and evaluation, then making every step visible through interactive, in-browser visualizations so you can actually see how an agent learns.
What you get
Features
Interactive RL Visualization
Real-time, web-based charts and views that make agent behavior and performance easy to read at a glance.
Automated RL (AutoRL)
Automate the tedious parts — algorithm selection, hyperparameter tuning, and evaluation — so you can focus on results.
Agent Leaderboard
Compare algorithms and configurations side by side, and filter by phase, epoch, iteration, and action in real time.
3D Edit Gym
Design environments in flexible TypeScript and watch agent trajectories play out in an interactive 3D scene.
Runs in Your Browser
Deploy in minutes with Docker Compose, then do everything from the web — no heavy local setup required.
Open-Source & Extensible
GPL-3.0 and built on ARLO & MushroomRL, integrating OpenAI Gym and Gymnasium for researchers and practitioners alike.
Product tour
See AutoRL X in action
Agent Leaderboard: Running a gym environment with different RL algorithms in AutoRL X's agent leaderboard. Users can select different agent configurations on the left and add them to the line chart to explore and compare their progress in real-time. In the menu above the chart users can filter by different phases, epochs, iterations, and actions of the RL model/agent, besides the option to select different tabs with agent logs, visited states, hyperparameters, and learned policies.
3D Visualization in Edit Gym: Users can create flexible TypeScript code to design a 3D environment that dynamically visualizes an agent's actions within AutoRL X. This example showcases agent trajectories within a simulated gym, providing improved interpretability and insights into RL agent behavior.
Get started
Installation
AutoRL X runs locally with Docker Compose, spinning up the database, server, and web interface with a single command. Prefer to hack on it? Run the Python server and Svelte frontend directly in developer mode.
git clone https://github.com/lorifranke/autorlx
cd autorlx
sh deploy_by_pull.sh # pull prebuilt images & launchTwo ways to run
- Docker (recommended) —
deploy_by_pull.shfor prebuilt images, ordeploy_by_build.shto build locally. - Developer mode — Python server on port
8000, Svelte + Vite web UI on port5173.
Full instructions and troubleshooting live in the GitHub repository.
About
About AutoRL X
AutoRL X is an open-source platform that makes reinforcement learning approachable. It brings real-time, web-based visualizations and intuitive interfaces to every stage of the RL workflow — bridging the gap between complex pipelines and the people who use them across healthcare, robotics, production, and autonomous systems.
Built on solid foundations
AutoRL X builds on ARLO and MushroomRL, with a Svelte frontend, and integrates industry-standard libraries such as OpenAI Gym and Gymnasium. It is released under the GPL-3.0 license.
Explore the GitHub repository or read the peer-reviewed ACM TIIS paper.
@article{franke2023autorl,
title = {AutoRL X: Automated Reinforcement Learning on the Web},
author = {Franke, Loraine and Weidele, Daniel Karl I and Dehmamy, Nima
and Ning, Lipeng and Haehn, Daniel},
journal = {ACM Transactions on Interactive Intelligent Systems},
year = {2023},
publisher = {ACM New York, NY}
}FAQ
Frequently Asked Questions
What is reinforcement learning visualization?
Reinforcement learning visualization turns an RL agent's training and behavior — its rewards, states, actions, and learned policies — into interactive charts and views, so you can understand, compare, and debug how the agent learns. AutoRL X does this in real time, directly in your browser.
What is AutoRL X?
AutoRL X is a free, open-source platform for Automated Reinforcement Learning (AutoRL) on the web. It lets you build gyms, run optimization experiments, and visually inspect agent performance — including an agent leaderboard and 3D environment views.
Is AutoRL X free and open source?
Yes. AutoRL X is released under the GPL-3.0 license, and the full source code is available on GitHub.
How do I install AutoRL X?
The quickest way is with Docker Compose: clone the repository and run
sh deploy_by_pull.sh to pull the prebuilt images and launch the database,
server, and web interface. Advanced users can run the Python server and Svelte frontend
directly in developer mode.
Which RL libraries and environments does AutoRL X support?
AutoRL X builds on the ARLO and MushroomRL frameworks and integrates industry-standard environments such as OpenAI Gym and Gymnasium.
What can I visualize with AutoRL X?
You can compare algorithms and configurations on an agent leaderboard, watch real-time training progress, and explore logs, visited states, hyperparameters, and learned policies — plus 3D agent trajectories in an editable gym environment.
Contact
Need Help?
For support, feel free to reach us at or submit an issue on our GitHub repository. For general inquiries, partnership opportunities, or collaboration requests, contact us at .