Precision in every pick.
Robotic component placement
3D vision helps robots locate, orient, and place non-standard electronic components.
- Industry
- Electronics manufacturing
- Our role
- 3D vision & robot guidance
- Core tools
- Structured light · YOLOv7 · RTX A4000
A practical problem.
Legacy template matching struggled with irregular connectors and reflective shielding cans. New components took days to configure, and the line needed more reliable positioning for a varied product mix.
- Engagement
- 7 months · Apr–Oct 2023
- Project setting
- UK engineering · Shenzhen production
Built around the constraints.
- See in three dimensions.
RGB and structured-light cameras capture component shape and position.
- Estimate the pose.
A trained detector and CAD-based refinement determine the orientation needed for placement.
- Pick, place, verify.
Grasp planning and visual checks connect perception to the existing assembly workflow.
How it comes together.
A vision-to-motion pipeline estimates a component’s pose, guides placement, and checks the result before the next cycle.
-
Capture
RGB cameras · Structured light
Synchronised images and depth measurements capture components in feeder trays. Calibration corrects camera distortion.
OutputColour images & 3D point clouds
-
Identify
YOLOv7 · RTX A4000
A trained detector locates and classifies components, including irregular connectors and shielding cans.
OutputComponent class & location
-
Orient
Learned keypoints · CAD registration
Keypoints initialise CAD-based pose refinement. Grasp planning selects a suitable vacuum-nozzle contact point.
Output6-DOF pose & grasp target
-
Place
Robot control · Visual verification
The robot aligns and places the component. Force feedback and a final visual check support error recovery.
OutputVerified placement or retry
Results from this project.
- Detection accuracy
- 98.8%Compared with 88% for template matching
- Placement accuracy
- ±0.05 mmWith controlled lighting and calibration
- Complete cycle
- 2.3 sPick, place, and verification
Before
Fixed templates struggled with irregular components and took days to configure for new parts.
After
3D perception guides pick-and-place, with new component calibration and validation reduced to around four hours.
In context. Transparent and highly reflective parts reduce detection accuracy to 94.2%. Controlled lighting and regular recalibration are needed; very small components require a different placement head.
Working on something similar?
Tell us about your system and where you need support.