> For the complete documentation index, see [llms.txt](https://docs.roboset.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.roboset.org/introduction.md). # Introduction {% hint style="info" %} Roboset is the data layer for Physical AI — a network for creating, verifying, and licensing the real-world data robots need to learn. Train policies in your browser, contribute task data through Atlas, and earn when your data trains real models. {% endhint %} ### Introduction Roboset is the data layer for Physical AI. It is a network for creating, verifying, and licensing the real-world interaction data that robots need to learn physical tasks — the data that barely exists anywhere else. The platform has two connected surfaces: The Training Platform (roboset.org) — a browser-based environment where anyone can train robot policies through simulation, reinforcement learning, and manual demonstration. Every attempt becomes structured data; every training run produces a saved, evaluable model; every useful contribution earns points. Atlas (atlas.roboset.org) — a data marketplace where contributors upload real-world task recordings or teleoperate robot arms, the network verifies and enriches them into datasets, and Physical AI labs license them. Contributors earn when their data is licensed. These docs serve four audiences: * Contributors who record data, run training, and earn rewards. * Buyers — Physical AI labs and robotics teams — licensing datasets or commissioning collection. * Developers integrating with the Roboset API. * Operators running nodes or hosting robot hardware as the network grows. Use the left navigation to jump to your path, or start with the Quickstart for your role below. ### Core Ideas in Five Minutes Physical AI needs two kinds of data. Human activity data — first-person video of people doing real tasks — teaches models what a task looks like and how it unfolds. Robot interaction data — structured observation-action-reward episodes from simulated and teleoperated arms — teaches models how those actions translate into control. Roboset collects both under one task taxonomy. Everything is an episode. Whether it comes from a human demonstration, an RL rollout, or a teleoperation session, the atomic unit of data on Roboset is the episode: a time-ordered sequence of observations, actions, rewards, and an outcome. Episodes are what get trained on, evaluated, packaged, and sold. Contribution is verified, not scraped. Every submission passes automated quality and privacy checks before entering the network. Buyers purchase from a verified pool, not a firehose. Points are the reward ledger. Contributors earn Roboset Points (RP) for useful work. RP is the accounting layer that determines each contributor's share of dataset revenue. RP is earned through verified work only — never purchased, never transferable. The people who make the data share in it. When a dataset sells, contributors are paid their share automatically, based on the RP behind the data that sold. ### Key Terms * Task — a defined manipulation objective (e.g. Pick Stone). Tasks have a difficulty level, a base reward, an example reference, and a backend environment. * Episode — one attempt at a task, recorded as observations, actions, rewards, and an outcome. * Demonstration — an episode produced by a human directly controlling the robot (manual) or by a guided expert routine (guided). * Policy — the neural-network model that decides what action to take given an observation. * Model run — a saved, trained policy with its metadata and metrics. * Dataset — a packaged, verified, enriched collection of episodes grouped by task, environment, and quality. * RP (Roboset Points) — the reward and revenue-share unit. * Contributor tier — a reputation level that unlocks higher-value tasks and reward multipliers. * Node — infrastructure that scrubs and processes data or hosts robot hardware. * Bounty — a buyer-funded request for a specific kind of data at a specified volume. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.roboset.org/introduction.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.