Lucky Engine¶
Lucky Engine is a robotics simulation engine: physics-driven scenes, real-time and path-traced rendering, simulated sensors, dataset recording, an editor, C# scripting, a gRPC server with a Python SDK for driving simulations from your own code, and an MCP server for AI assistants.
These docs cover the editor, its simulation systems and the scripting API, one section per
area. Use the version selector in the header to switch between engine releases (for
example 2026.2 ↔ 2026.3).
Docs for the version you run
Help → Documentation and the side bar's Open the Documentation button open these docs at the version of the editor you're running.
Get started with a tutorial¶
See more example projects and tutorials.
Getting started¶
-
Release highlights for Lucky Engine 2026.3, the current stable release, and for earlier releases.
-
Install Lucky Engine, find your way around the Welcome screen, sign in, run the Welcome scene, and create a first project.
-
Ready-to-run projects: the bundled Welcome marble run, and examples from the Content Vault such as Piper Pattern Stacking and Walking Pick and Place.
-
Record demonstrations of an SO-100 arm in Lucky Engine, train an ACT policy on them, and deploy it to the real robot.
Using the editor¶
-
The title bar, side bar, dock zones, status bar and notifications, the menus, and every panel and how to open it.
-
The built-in Scene, Files, Server and Observe layouts, your own custom layouts, tabs and splits, separate windows, and Fullscreen.
-
The Scene viewport and its toolbar, adding content to a scene, and the Play-mode tools. Viewport Navigation covers moving the camera and the transform tools.
-
The Content Browser and the project's Assets folder: importing source files, creating assets, and dragging them into a scene.
-
Browse and download robots, environments, assets and example projects, and keep installed packs up to date.
Simulation¶
-
Robot packs, placing a robot in a scene, the Robot Controller, and adding policies and motion graphs.
-
Graph assets for procedural geometry, robot motion, animation and audio. Build them in the graph editor, expose parameters, and tune each placed instance.
-
The Rasterized and Path Traced render modes, denoisers and DLSS, cameras and exposure, lights and materials, and the Scene Renderer panel.
-
Simulated lidar and RGBD depth: scan patterns and presets, the lidar return model, the depth sensor model, lidar replay, and reading both from script or over gRPC.
-
Record datasets from a running scene: the Observer panel, camera and lidar bindings, video encoding on the GPU or with FFmpeg, and the dataset format on disk.
-
The fixed-timestep orchestrator: up to eight runners on one shared timeline, and five execution phases per simulation step.
-
SI units throughout: metres, seconds, kilograms, newtons. Radians in the script API, degrees in the Inspector, and where each quantity shows up.
Code and automation¶
-
Write C# to drive behaviour in your scenes: move entities, respond to input and collisions, query physics, control robots, and define reinforcement-learning tasks. Includes a getting-started path, task-focused guides, and the full API reference.
-
The engine's gRPC server: drive simulations from any language. Step the agent loop, read and write MuJoCo state, control policies, and negotiate RL tasks. Includes a Python helper.
-
Connect an AI assistant to the editor's built-in MCP server, which runs on your computer whenever the editor is open.
-
Launch options for automation: open a project and scene, start the gRPC server on launch, and choose the MCP server's ports.