Fixed mount
Above the bin
- Captures while the robot moves, so vision adds no cycle time
- One view covers the whole bin
- No camera cable on the robot arm
◆ AI-powered 3D vision
Eureka
3D Camera
Our range of AI Stereo Vision cameras designed for 3D picking delivers unmatched accuracy in challenging production conditions.
01 — AI StereoVision
AI-powered part detection with
99.5% picking accuracy
High-resolution point clouds give the AI a precise view of every part, for reliable identification and picking.
- 1Shiny or translucent partsReflective metal and see-through plastic.
- 2Overlap and bin cornersParts stacked on each other or pressed into the corners of the bin.
- 3Parts that differ from CADThe physical part is not exactly as the CAD file.
1.6 MP
Global shutter, up to 1440 × 1080
No laser
Passive AI stereo, no laser needed
0.2s
Vision cycle, as low as
0s
Effective vision cycle when run parallel to robot motion
02 — Software
Program your 3D vision cell with StereoPick
Intuitive software for 3D picking applications.
Explore StereoPick
03 — Models and specs
Model ranges
Download user manual- Working distance
- 300–600 mm
- FOV at min. distance
- 210 × 188 mm
- FOV at max. distance
- 448 × 376 mm
- Lateral resolution
- 0.17–0.35 mm/px
- Expected depth error (calculated)
- 0.26–1.04 mm
- Baseline
- 120 mm
- Size (L × D × H) · weight
- 160 × 65 × 43 mm · 0.6 kg
| Model | Close range ECA-S-3001 | Mid range ECA-M-3002 | Long range ECA-L-3002 |
|---|---|---|---|
| Working distance | 300–600 mm | 600–1200 mm | 1200–4000 mm |
| FOV at min. distance | 210 × 188 mm | 454 × 374 mm | 917 × 746 mm |
| FOV at max. distance | 448 × 376 mm | 918 × 749 mm | 3186 × 2486 mm |
| Lateral resolution | 0.17–0.35 mm/px | 0.35–0.69 mm/px | 0.69–2.3 mm/px |
| Expected depth error (calculated) | 0.26–1.04 mm | 0.80–3.18 mm | 3.18–35.38 mm |
| Baseline | 120 mm | 156 mm | 156 mm |
| Size (L × D × H) · weight | 160 × 65 × 43 mm · 0.6 kg | 200 × 65 × 45 mm · 0.9 kg | 200 × 65 × 45 mm · 0.9 kg |
All models
- Sensor
- CMOS colour, 1.6 MP, global shutter
- Resolution · capture time
- Up to 1440 × 1080 · 0.1–1.2 s typical
- Depth
- AI stereo matching, passive (no laser)
- Outputs
- Colour image, depth map, point cloud, stereo pair
- Interface · power
- GigE Vision, M12 X-coded · PoE 802.3af, max 15 W
- I/O · trigger
- M12 8-pin: 1 in, 1 out, 2 GPIO · software, hardware
- Environment
- 0–50 °C · IP55 (test pending)
- In the box
- Mounting bracket, calibration set, 5 m M12–RJ45 cable
Fanuc, ABB, UR, AUBO out of the box · open APIEvery model: fixed or robot-mountedOne PoE cable: power and data
04 — Mounting
Mounts to the robot or above the bin
Robot-mounted
On the robot hand
- Higher precision from a close-up view
- Looks at parts from several angles to find more picks
- Pauses the robot while it captures
Plain, textureless parts?
Add the optional pattern projector. It gives the stereo pair texture to match on, with no laser.
Compatible with
all major robot brands
Your robot program asks for the next pick and gets back where to grab it. Everything runs on-site on the Eureka Controller, with no cloud connection and no subscription.
Integrated out of the box
Any other robotvia open API
05 — Comparison
AI Stereo vs. other 3D methods
Depth from
- AI Stereo
- Two images, matched by AI
- Structured light
- Projected light patterns
- Time-of-flight
- Travel time of emitted light
Shiny parts
- AI Stereo
- Handled well
- Structured light
- Glare leaves gaps in the data
- Time-of-flight
- Reflections distort depth
Transparent parts
- AI Stereo
- Some today, improving with updates
- Structured light
- Pattern scatters
- Time-of-flight
- Light passes through
Matte, opaque parts
- AI Stereo
- Accurate enough for most picking
- Structured light
- Most accurate (sub-mm)
- Time-of-flight
- Lower depth detail
Gets better over time
- AI Stereo
- Yes, AI models improve with software updates
- Structured light
- Fixed by the optics
- Time-of-flight
- Fixed by the sensor
AI Stereo | Structured light | Time-of-flight | |
|---|---|---|---|
| Depth from | Two images, matched by AI | Projected light patterns | Travel time of emitted light |
| Shiny parts | Handled well | Glare leaves gaps in the data | Reflections distort depth |
| Transparent parts | Some today, improving with updates | Pattern scatters | Light passes through |
| Matte, opaque parts | Accurate enough for most picking | Most accurate (sub-mm) | Lower depth detail |
| Gets better over time | Yes, AI models improve with software updates | Fixed by the optics | Fixed by the sensor |
Find an integrator
Work with a Eureka-trained integrator in your region to deliver the full cell.