Release Notes¶
Patch history for this release line. Each heading is a minor patch and is directly linkable, e.g. 2026.3. For the feature highlights of a release, see What's New.
2026.3¶
14 October 2026
Stable release. See What's New for the full feature list.
- Added a Path Traced render mode next to Rasterized, switched with the viewport toolbar's Rasterized and Path Traced buttons or in the Scene Renderer; it needs a GPU with hardware ray tracing on Windows or Linux
- Added path-tracing denoisers: ReBLUR, ReLAX and, on NVIDIA RTX GPUs, NVIDIA DLSS Ray Reconstruction with Quality, Balanced, Performance, Ultra Performance and DLAA modes
- Added procedural graphs (
.hpodgraph) for procedural geometry, robot motion, animation and audio, edited in a graph editor that opens as a tab; create them from the Content Browser's New → Procedural Graph menu - Changed robot motion graphs, animation graphs and sound graphs to procedural graphs; New →
Motion Graph and New → Animation Graph now create
.hpodgraphfiles - Added a side bar with the built-in Scene, Files, Server and Observe layouts, plus Home, Report a Bug, Open the Documentation, Settings and Profile buttons; see Editor Interface
- Added custom layouts: create, rename, re-icon and reorder them,
save them to the project or to
.hlayoutfiles, and load them in other projects - Added a new Welcome screen with recent projects, templates (New Project, Unitree G1, Emika Panda) and example downloads; see Get Started
- Added multiple Scene viewports, each with its own camera and settings: open another from the Main zone's add-panel menu, split it beside the first, or move it to its own window
- Added path-traced recording per camera, with a Denoiser choice (including OIDN and NVIDIA DLSS Ray Reconstruction), Quality presets, Bounces, Samples and Sub-frames
- Added the Custom Schema recorder mode: list the MuJoCo values to record in a YAML file and get one typed Parquet column for each; see Custom Recording Columns
- Added in-process GPU video encoding (H.264 and HEVC) on Windows and Linux, the default Video Format for new scenes, so colour video records without the external FFmpeg program on supported GPUs
- Added near-infrared material channels for glass, water, mirrors and foliage, with twelve measured 905 nm presets
- Added per-sensor Secondary Capture for secondary returns: an opt-in second hit behind glass, water or foliage, recorded as material-agnostic geometry
- Added progressive refinement and adaptive sampling for clean path-traced stills, plus Indirect bounces, Bright sample limit and Environment multiplier settings
- Added camera exposure modes: Manual, Manual (Physical Camera) and Automatic (Eye Adaptation)
- Added a Physical mode to the Camera component, with lens, aperture and sensor settings for path-traced rendering
- Added area lights and emissive materials as light sources in Path Traced mode, with soft shadows; see Lights and Materials
- Added field of view, clip planes, exposure modes and, in Path Traced mode, depth of field with focus distance, sensor height and vignetting to the viewport camera, set from the toolbar's Camera menu
- Added path-traced view modes for inspecting geometry, materials and motion, such as Normals, UVs, Albedo and Surface Motion
- Added NVIDIA DLSS upscaling for the Rasterized mode on NVIDIA RTX GPUs, in the Scene Renderer under Rendering → Upscaling
- Added a Convolution (FFT) bloom type; bloom and depth of field work in both render modes
- Added a Directional shadow bias setting to reduce shadow acne
- Added procedural primitives (Create → 3D → Procedural Cube, Procedural Sphere and seven more) and the Procedural Mesh component; Promote to Graph turns a primitive into an editable graph
- Added exposed graph parameters to the Inspector, which each instance can override and restore with Reset to Default
- Added MuJoCo collision for procedural geometry (Output Role, optional Convex Decomposition); graphs with several bodies and joints become articulated bodies automatically
- Added cables built by procedural graphs, which bend and hang under MuJoCo physics; see Cables
- Added
PodComponentmethods for changing graph parameters from scripts (SetParameter,ResetParameter) and for making generated bodies glow (SetBodyMaterialEmission,ResetBodyMaterialEmission,ClearMaterialOverrides) - Added a Procedural Graphs category to the Content Vault
- Added separate windows: drag a tab out of the editor, or choose Move to new window from its right-click menu; View → Reopen Last Closed Window brings a closed one back
- Added a Fullscreen button at the right of the main tab strip that hides the side and bottom zones, also toggled with Tab in a focused Scene viewport
- Added title-bar buttons for File Tree, Scene Hierarchy, gRPC Server, Content Browser and Logs, replacing the panel buttons over the viewport
- Added a status bar with the current scene, the editor version and a notification centre; messages such as Project Saved. now appear as notifications
- Added Help → Check for Updates, and a notification when a newer version is available
- Added new number fields: drag to change a value (Shift for finer,
Ctrl for coarser steps), click to type, or enter arithmetic such as
0.5*3orsqrt(2) - Added axis badges to vector fields: click X, Y or Z to reset that component
- Added Tab and Shift+Tab to move between number fields, and unit suffixes such as
cm,degandmsin the viewport toolbar's number fields; see Editing Values - Added one unsaved-changes prompt for the scene, Text Editor tabs and graph editors, shown when you close a tab, switch scene or project, or quit; tabs with unsaved edits show a dot
- Added Image Viewer, Video Player and Parquet Table tabs: images, videos and recorded datasets open in their own tab, one per file
- Added UI scale shortcuts: Ctrl+= and Ctrl+- change the scale, Ctrl+0 resets it to 100%; see UI scale
- Added a toolbar to each viewport with the transform tools, Play controls, Recorder Status, the Rasterized and Path Traced buttons with a Path Tracing menu, Overlay and Colliders toggles with their menus, and a Camera menu
- Added per-tab viewport settings: each Scene tab remembers its own camera, overlays and render mode, and Set Scene Default View stores a tab's view in the scene; see What each Scene tab remembers
- Added a Robots panel (View → Robots) for finding, downloading and adding robots, also opened from Robots in the viewport's Create menu
- Added Duplicate to the viewport's right-click menu
- Added 3D icons for cameras and spot lights that show the camera's frustum or the light's cone when selected, and Viewport Icon settings (Visible, Size) on the Camera component
- Added multi-asset drops: drag several meshes, static meshes or procedural graphs into the viewport at once and they are laid out side by side; several mesh sources open their import wizards in turn
- Added an Up to parent directory button to the Content Browser; Back and Forward now follow your browsing history
- Added a new searchable picker for Add Component and for asset, entity and script fields in the Inspector, with keyboard navigation and a list that stays current as assets are added or removed; see Reference fields
- Added Depth Mode, Frame Rate (24 to 120 fps) and Render Mode settings to each camera binding; see Recording
- Added on/off switches for the data, camera and lidar recorders and for each camera, with a status line per camera and Open Output Folder, to Recorder Status
- Added
neural_denoiseandbit_reproducibletometa/info.json; a dataset recorded with NVIDIA DLSS Ray Reconstruction is marked not bit-reproducible; see Dataset Format - Added
Observer.IsEpisodeOpen; see Recording with the Observer - Added nine diagnostic channels per depth camera, reporting one internal sensor-model term per pixel
- Added stale-bake detection to the Lidar Replay panel: a bake that no longer matches the scene's materials or bake settings, or comes from an earlier version, offers Re-bake and blocks export until you re-bake
- Added a Calibrated reflectivity point colour mode for lidar
- Added path-traced camera frames over gRPC:
Stepcamera requests,StreamCameraandStreamViewportreturn path-traced colour for cameras whose Recorder binding uses Path Traced - Added per-beam intensity and calibrated reflectivity to
GetLidarScan, plusGetLidarReturnParams,GetLidarBakeStatusandSetLidarSecondaryCapture; see Sensors over gRPC - Added
depth_modeanddebug_channelto the gRPC camera configuration - Added lidar and depth methods to
MujocoSceneComponent, includingGetLidarScanChannels,GetLidarBakeStatus,SetLidarSecondaryCaptureandSetCameraDepthDebugChannel - Added MCP tools for creating projects from Content Vault packs, authoring procedural graphs and building scenes from them
- Added the
--helpand--versioncommand-line options - Changed the Robot Controller's Motion Graph field to take a procedural graph and list the graph's inputs for per-robot editing (right-click for Use graph default or Send pulse); see Attaching a motion graph
- Changed the Sound Config Source to accept an audio graph as well as an audio file
- Changed the bundled starter project to Welcome, in which a Panda arm picks marbles out of a marble run built from procedural graphs, set to Path Traced; see Example Projects
- Changed new scenes to default to Path Traced, and projects to use NVIDIA DLSS Ray Reconstruction as the denoiser where the GPU supports it; scenes from earlier versions open Rasterized
- Changed cameras to default to Automatic (Eye Adaptation) exposure, including cameras in scenes from earlier versions
- Changed point and spot light falloff to be physically based, and removed their Falloff setting
- Changed the Material Editor's Glass section to IOR, Attenuation, Dispersion and Tint; a Transparent material is always glass, and the Transparency, Transmission and Thickness controls are gone
- Changed the Scene Renderer panel: a new Rendering page, the SSR page renamed Reflections, and clearer labels and tooltips
- Changed ambient occlusion to darken only ambient lighting, so directly lit edges are no longer over-darkened
- Changed depth recording to be set per camera: each camera binding has its own Depth Mode, and Default Depth Mode only seeds new bindings; scenes from earlier versions migrate automatically
- Changed the depth sensor model to read the authored near-infrared material channels, so depth sees through glass and water at the right strength
- Changed lidar bake settings to be saved with the scene
- Changed Recorder capture settings to be editable only while the scene is stopped and no recording session is armed
- Changed Record Video into two settings, Record Cameras and Record Color Video; see Recording
- Changed a missing FFmpeg program to no longer switch recording off: the cameras that need it show Needs FFmpeg, with Install FFmpeg and Use In-Process GPU (H.264) offered. Depth, masks, lidar video proxies, the FFmpeg formats and video playback still need it; see Video Encoding and FFmpeg
- Changed sign-in to the side bar's Profile button
- Changed the editor's executable name from
LuckyEditortoLuckyEngine; see Command Line - Changed the Content Vault to list packs in the order they are published instead of by name
- Changed Audio Events to open from the View menu, and moved the keyboard shortcut reference from the Help menu to Settings (Application → Shortcuts)
- Changed asset editors such as Prefab, Audio File and the MuJoCo Import Editor to open as tabs instead of floating windows
- Changed the gRPC Server panel to open from the title bar, the side bar's Server layout or View → gRPC Server
- Changed C# script projects to be generated from a
premake5.luain the project that you can customise; see Creating and Building Scripts - Changed
MujocoSceneComponent.SetCameraConfigto takeint depthMode(0 Off, 1 RGBD, 2 Separate, 3 DepthOnly) instead ofbool recordDepth;GetCameraConfigreturns the depth mode where it returned the flag - Changed
Observer.IsRecordingto report the scene's actual recording state, including sessions not started from a script; see Recording with the Observer - Changed
Stepso that a camera request that cannot be captured no longer blocks the step: the simulation advances and the observation is returned, withsuccess=falseand a message - Changed
SetLidarReturnParamsover gRPC to accept optional optics fields; optics fields a request omits keep their current values - Changed the MCP
spawn_robottool to install a missing robot pack in the background and return without placing the robot, andcreate_sceneandopen_sceneto refuse to discard unsaved changes; see AI Assistants (MCP) - Changed command-line handling: unknown or repeated options, more than one project path, a
missing project file and
--headlessnow stop the editor with an error instead of being ignored; see Command Line - Removed the Simple and Advanced editor modes; there is one editor interface, arranged with layouts
- Removed the in-editor Agent panel; AI assistants connect through the editor's MCP server
- Removed the legacy motion graph (
.hmograph), animation graph (.hanimgraph) and SoundGraph (.sound_graph) formats and their editors, including New → Audio → SoundGraph Sound; packs and imported projects that contain them are rejected, and scenes that use one on a Robot Controller or an Animation component no longer open; see Upgrading from 2026.2 - Removed the Simulate Physics button, and the scene drop-down and Build button in the title bar; use File → Recent Scenes and Build → Build C# Assembly instead
- Removed the Data Viewer panel;
.parquetfiles open in a Parquet Table tab - Removed the scene-wide Depth Mode and the per-camera Record Depth checkbox; use each camera binding's Depth Mode
- Removed dragging robots from the Content Vault panel into the scene; place robots from the Robots panel or Robots in the viewport's Create menu instead; see Robots
- Removed the MuJoCo Force Tool panel; its settings are in the menu beside Force Tool on the viewport toolbar during Play; see The force tool
- Removed Animation, Audio, Audio Listener, Sprite Renderer, Tile Renderer, Source Reflection, Room Space and Late Reverb from the Add Component menu
- Improved recording robustness: dataset metadata is saved every 25 episodes during the session, so an interrupted session keeps the episodes up to the last save, and an episode whose video, depth or lidar stream fails is marked unsuccessful
- Improved script project upgrades: an incompatible engine-written
premake5.luafrom an older release is kept aspremake5.lua.legacyand replaced automatically - Improved the Content Vault and Welcome screen example lists, which now appear immediately from a local cache and then refresh
- Improved Content Browser file-type icons, with thumbnails for procedural graphs
- Improved camera navigation to accept the right-hand Shift, Ctrl and Alt keys
- Fixed the Directional shadow distance (m) setting cutting off all lighting beyond it; it now fades only the shadows
- Fixed the farthest shadow cascade ending short of the shadow distance
- Fixed specular banding and colour shifts on very bright highlights
- Fixed FBX and GLB models that reuse one mesh in several places showing only one copy
- Fixed the External Action Gate re-arming after a gRPC client
reports it has finished (
ReportProgresswithfinished=true); the scene keeps running until the client callsSteporResetAgentagain - Fixed Internal Phase Priority having no effect on the order of systems within a phase; see Order within a phase
- Fixed the Sync Point System losing its 10 Hz runner when a scene was saved
- Fixed crashes when an MCP client accessed the editor, and undo for scene edits made through MCP tools
2026.2¶
21 July 2026
Preview release. See What's New for the full feature list.
- Added LiDAR sensors, with viewport preview, Livox non-repetitive scan patterns, and per-sensor recording
- Added depth and RGBD recording, written to the dataset as
observation.images.<Name>_depth - Added per-camera depth scale, which fixes near-range depth reading as black
- Added a
LidarServiceto the gRPC API, plus depth frames onStep - Added a material-informed export to the Lidar Replay panel, with
an
augmentation.jsonrecording each variant's provenance - Changed the renderer to a new graphics backend
- Changed editor icons to stay sharp at any size and on high-DPI displays
- Improved stability when opening and switching projects
- Fixed a long-standing set of shutdown and project-switch crashes
2026.1.2¶
21 July 2026
Minor update with fixes and improvements.
- Added SO100 Pick And Place example scene to the Content Vault, used by the LeRobot: Sim-to-Real Tutorial
- Added engine version requirements to Content Vault packs
- Added a dimmed "Requires" label in place of Download when a pack needs a different editor version
- Fixed SO-100 gripper force and damping overrides
- Fixed Viewport Settings dropdowns vanishing behind the editor on some systems
2026.1.1¶
19 June 2026
Minor update with fixes and improvements.
- Added MuJoCo Import Editor (double-click on MuJoCo XML in Content Browser)
- Added better .NET version detection and user feedback
- Added more macOS ffmpeg search paths, eg. homebrew
- Added API Reset Settle Steps to Scene Settings
- Changed API function/enum names and docs to encourage AI agents to reason better
- Improved MuJoCo XML support and import stability
- Fixed vsync using incorrect presentation mode resulting in higher than necessary GPU usage
- Fixed Agent panel always being open on startup
- macOS crash fixes:
- Fixed startup crash due to conflicting Vulkan/MoltenVK loaders
- Fixed crash when resizing window
2026.1¶
12 June 2026
The first public release. See What's New for the full feature list.