Repository navigation
Releases: microai-lab/microduck-studio
Releases · microai-lab/microduck-studio
Release list
Microduck Studio v0.3.0
Highlights
- Added live MuJoCo world and head-camera views.
- Added selectable simulator scenes with safe restart orchestration.
- Expanded robot telemetry with trunk/head IMU, ToF depth, joint errors, path history, and loop-rate monitoring.
- Added unified status and restart controls for robotd, tofd, and MuJoCo.
- Added configurable visualization quality profiles and camera controls.
- Added an end-to-end contract probe covering runtime source, policies, sensors, movement, stop, and disconnect safety.
- Improved native macOS background rendering and Docker-based development workflows.
Compatible components
- microduck:
studio-microduck-v0.3.0 - microduck_rl:
studio/microduck_rl/v0.3.0
Microduck Studio v0.2.0
New Features
- Added an authoritative MuJoCo scene view in the browser, with orbit, zoom, and camera reset controls.
- Added a comprehensive real-time
robotdtelemetry dashboard, including:- Active policy and control commands
- IMU, odometry, and power status
- Measured and commanded joint values with tracking errors
- Control-loop frequency, robot thumbnail, and movement path
- Added safe service start and restart controls for managing
robotdand MuJoCo from Studio. - Added native background MuJoCo mode for macOS, with an optional desktop Viewer mode.
- Added Docker-based MuJoCo modes with support for:
- CPU/OSMesa rendering
- Linux DRI/EGL GPU acceleration
- NVIDIA/EGL GPU acceleration
- Added an end-to-end control probe to the unified launcher, verifying that the complete Studio →
robotd→ MuJoCo control path is operational. - Added configurable rendering resolution, frame rate, and image quality.
- Improved the control interface for both mobile and desktop devices.
- Added Chinese and English interface switching, and fixed the training task input placeholder so that it updates with the selected language.
Operations and Safety
- Introduced a restricted host service manager that accepts only predefined start and restart operations.
- MuJoCo restarts now wait for the simulator to recover and then restart
robotdto restore the control chain. - Improved process and container cleanup when switching between runtime modes.
- Added an option to stop background MuJoCo after the last browser page has remained disconnected for a configurable grace period.
- Preserved Studio’s role as an observability and orchestration layer without bypassing
robotdcontrol or safety boundaries.
Documentation
- Expanded the English and Chinese documentation for installation, unified startup, rendering modes, troubleshooting, and architecture.
- Added a real Studio runtime screenshot to the top of the README.
Full Changelog: v0.1.0...v0.2.0
Microduck Studio v0.1.0
Microduck Studio v0.1.0
Microduck Studio is a local developer console that connects the microduck
robot runtime with microduck_rl simulation and training tools.
This initial release brings browser-based control, runtime visibility, and
repeatable startup into one development workflow.
Highlights
- Bilingual Web controls with Chinese labels and English subtitles.
- Movement, enable/stop, and skill commands through the existing robotd interface.
- Runtime, MuJoCo, repository, policy, and job status in one place.
- One-command macOS startup for Docker-based robotd and Studio, plus a native MuJoCo Viewer.
- End-to-end startup verification that confirms control commands move the simulated robot.
- Direct access to
robotctl monitorthrough a disposable Docker Compose container. - Optional, allowlisted RL smoke tests, disabled by default.
- English and Chinese setup guides, including prompts for local AI coding agents.
Project boundaries
microduckowns runtime control, policy inference, hardware I/O, and safety.microduck_rlowns MuJoCo simulation, RL training, and policy export.microduck-studioowns the Web interface, status aggregation, and local orchestration.
Requirements and limitations
- The complete development stack requires all three repositories in sibling directories.
- The one-command launcher currently targets macOS and requires Docker with Compose.
- MuJoCo integration uses the
sim-remote-ioruntime branch. - Studio has no authentication; keep it on localhost or a trusted network.
- Backward walking depends on the loaded policy and is not guaranteed by this release.
Getting started
Follow the README to prepare the repositories and dependencies, then run:
./scripts/dev-stack.sh
Wait for control probe passed before using the Web controls.
Full Changelog: https://github.com/microai-lab/microduck-studio/commits/v0.1.0