BotManifold Docs
BotManifold is a workbench for robot teams: build scenarios, simulate in the browser, verify policies, teleoperate hardware, and manage datasets.
BotManifold is a workbench for robot teams. You build a scenario, run it in a physics simulator in your browser, drive it with a policy, teleoperate real hardware, store the data, and get a graded safety report for a policy. It replaces the separate simulation, data and test tools a team would otherwise build and keep up.
Products
| Product | What it does | Docs |
|---|---|---|
| Studio | Browser scenario editor and MuJoCo simulator with an asset library, policy playback, drive mode and an AI copilot | Studio |
| Verify | Runs your policy through seeded simulation episodes and returns a verdict and a scored report | Verify |
| BotCI | Runs Verify from GitHub Actions so a failing policy fails the workflow | BotCI |
| Teleop and hardware | Drive and record a real arm (SO-100, myCobot 280) from Studio, locally or remotely, and run hardware diagnostics | Teleop |
| Datasets | Upload recorded data, process it into the LeRobot v3 format, and browse episodes and signals | Datasets |
| SDK and CLI | Python client and the botmanifold command | SDK |
| REST API | The HTTP API behind all of the above | API |
Where to start
- Check a policy for safety: Getting started runs your first verification in a few minutes.
- Try a policy in the browser: open Studio, add a robot, and play one of the example policies.
- Drive a real arm: follow the Teleop guide.
- Learn how modern robot-learning models work: the Tutorial builds up from transformers to vision-language-action models for driving, with quizzes and runnable PyTorch code.
- Know what a scenario expects: the scenario reference lists the observations, actions and success criteria for each one.
Reference
- Scenarios: the available tasks
- Writing policies and debugging tips
- FAQ