🤖
Ecovacs BaJie
6-axis collaborative arm + lift column + two-finger gripper; control, 3D digital twin, vision/voice, data & training, test & acceptance — fully open.
Open-source modelsorganised by brand / model
➕
More brands / models
More robot brands and models will be added and archived per model for easy lookup.
🦾
Hardware
Ecovacs · BaJie: 6-axis collaborative arm + lift column (0–0.44 m) + two-finger gripper (0–0.69), head and wrist cameras.
🧠
Control stack
Jetson → adb → chroot → ssh → ROS 2: joint trajectories, streaming jog, collision guards, auto real/local mode.
👁️
Perception & HMI
Head / wrist camera streams, object detection, voice commands and TTS, mission command entry.
🔁
Data loop
Collect → annotate → limits → training analysis → acceptance, with a tool and data format for each step.
Control5 modules · operation layer
From joint execution to 3D twin, gripper, voice and missions🎛️
Main console
Live dashboard, dev console, robot app, joint execution, lift and gripper control; motion log and state monitoring.
Source
host/index.html · host/joints_bridge.py🤖
3D arm
Real meshes, drag teaching, fixed-point head motion, live limit guard, wave mode; zoom and fixed camera pose.
Source
host/arm3d.html✊
Gripper control
Opening slider, close / full-close / open actions, live feedback; drag and one-tap motions.
Source
host/grip.html🎙️
Voice control
Voice commands, TTS announcements, microphone state; turn frequent motions into one sentence.
Source
host/voice.html · host/voice_server.py🕹️
Mission command
Robot command / mission entry with unified dispatch and status feedback.
Source
host/cmd.htmlData & Training4 modules · data loop
Collect → annotate → limits → analyse; make motions reproducible📊
Training data analysis
Task filters, quality metrics, action Δp99, episode replay and HDF5 dataset browsing — see if a sample is usable.
Source
host/data_viewer.html🎥
Fixed-point data collection
3D preview, real/sim fixed-point head motion, live state, head & wrist cameras, episode marking and saving.
Source
host/headfix_collect.html📐
Limit database
Critical tip angles, direction coverage, SQLite records and CSV export — quantify table/column boundaries.
Source
host/headfix_limits.html · host/limits_db.py🏷️
Annotation
Annotate grasp objects and placement locations with head & wrist camera views; produce a trainable dataset.
Source
host/annotate.htmlTest & Acceptance1 system · reliability
From “it moves” to “it is reliable”✅
Test plan
Motion / calibration / perception / data / training / reliability checklists and progress — the acceptance baseline.
Source
host/test_plan.html · docs/test_plan.jsonTools3 tools · authoring & debug
Motion authoring and camera debug💃
Dance editor
Keyframe authoring, 3D preview, import/export and real-robot playback — turn a sequence into a reusable show.
Source
host/dance_editor.html📷
Head camera
Head detection stream (percd, :8799) as live MJPEG for global view and target finding.
Source
host/camfeed.py · lower/jogd/percd.py🎯
Wrist camera
Wrist / gripper detection stream (percg, :8800) as live MJPEG for fine alignment and grasp confirmation.
Source
host/camfeed.py · lower/jogd/percg.pyGet control accessEcovacs BaJie robot
The control interfaces of the BaJie arm (mvp03 6-axis arm + lift column + head pitch + gripper) are available to lab-network members and authorised partners. Apply using the steps below; after approval you receive device addresses, accounts and console access.
- Eligibility: join the OpenHW Lab local-lab network, or own / borrow a BaJie device.
- Apply: email your organisation / contact / device model / use case / network environment.
- Review & authorise: we confirm the use case and safety conditions; access to the Jetson console, robot LAN and lower-board SSH is granted as needed.
- Safety briefing: read and confirm the motion safety rules (E-stop, limits, grasp rules).
- Docs & integration: download the API reference; we support the integration.
BaJie Robot Control API v1.0
Three access paths (HTTP API / Mission WebSocket / ROS 2), ports and command tables, safety limits, deployment checklist and troubleshooting.
Download API reference (.md · 22 KB, Chinese)
IPs and passwords in the document are placeholders and provided separately after authorisation. Interfaces have no auth by default; use only on a trusted LAN.
Three access paths (HTTP API / Mission WebSocket / ROS 2), ports and command tables, safety limits, deployment checklist and troubleshooting.
Download API reference (.md · 22 KB, Chinese)
IPs and passwords in the document are placeholders and provided separately after authorisation. Interfaces have no auth by default; use only on a trusted LAN.
Document SHA-256
e4b6a712ccce432279b6fdc3c95b0c986a11256db0105fd54346922ab483e1adDownload & Quick StartMIT · source package
📦
Ecovacs BaJie robot source package
Host console + lower-board scripts + meshes + docs — 216 files.
SHA-256
fc360d3670869f239f4b80a0ee27a1f30d3d9048f1a902b398b3f3fb04601d5f
Three steps:
- Unzip and edit paths plus
ROBOT_HOST/ROBOT_PWinhost/serve.sh. - On the host run
bash host/serve.sh, openhttp://<jetson-ip>:8790/robot_home.html. - Put
lower/jogdon the lower board and startjtstream.pyto connect ROS 2.
No robot? You can still open arm3d.html and other pages with a local static server to explore the 3D model and UI.
⚠️ Safety: the arm is a powerful real machine. Clear the workspace, keep the E-stop in hand, and validate unloaded at slow speed first. Always approach grasps from above. See docs/MOTION_RULES.md and docs/GRASP_RULES.md.
Verify & Version
Verify: Windows (PowerShell)
Changelog: 2026-10-24 · first public release (216 files / MIT): console + 3D digital twin + data collection/training analysis + ROS 2 lower-board scripts.
Get-FileHash .\openhwlab-ecovacs-bajie-robot.zip -Algorithm SHA256; macOS / Linux sha256sum openhwlab-ecovacs-bajie-robot.zip. Matches the SHA-256 above → intact official package.Changelog: 2026-10-24 · first public release (216 files / MIT): console + 3D digital twin + data collection/training analysis + ROS 2 lower-board scripts.