Jumbled stacking, mutual occlusion, smooth featureless surfaces—bin picking is a recognized shop-floor hard problem. With DaoAI 3D vision, a parts plant let robots understand each part's spatial pose, handing the most tedious loading task to machines.
Loading looks simple but is a long-standing automation headache. This plant's metal structural parts sat loosely piled in bins—random poses, layered occlusion, and smooth reflective surfaces lacking matchable texture features. Traditional 2D vision yields only planar position and cannot judge a part's flip or tilt, so robots frequently slipped, hit the bin, or grabbed nothing. Automation attempts stalled repeatedly, leaving workers to bend and hand-pick parts under heavy strain and unstable cycle times.
From Planar Coordinates to 6D Pose
DaoAI's in-house 3D camera images the whole bin in a single shot to generate a dense point cloud. Algorithms powered by the DaoAI World model estimate each part's 6D pose (3D position plus 3D orientation) directly, identifying parts reliably even under featureless, highly reflective conditions. The system then plans collision-free pick paths and picking order, taking parts layer by layer from top to bottom for true random picking.
- 99%+ pick success: covers reflective, occluded and wall-hugging hard poses
- 4–8s cycle per part: imaging, recognition and path planning all done inline
- No feature points needed: relies on 3D shape rather than texture, suited to smooth metal parts
- Collision-aware planning: auto-avoids bin walls and adjacent parts, protecting gripper and workpiece
The robot no longer grabs blind; it first understands each part's true pose in space.
After go-live, this station achieved unmanned loading. The stable high success rate kept downstream equipment from starving on empty grabs, made overall cycle time controllable, and fully freed workers from repetitive bending and hand-picking.