Model matching
Visible surfaces are associated with known geometry. Variation, damage and limited visibility affect the fit.
Knowledge · Robotics
Pose estimation: from 3D measurement to robot coordinates
A robot needs more than confirmation that a part was found. Pose estimation calculates its three-dimensional position and orientation, determines whether the result is reliable enough and transforms it into the robot's coordinate frame.
Six degrees of freedom
Position and orientation belong together
A 6D pose consists of three translations, X, Y and Z, and three rotations. Software may compare a point cloud or depth image with a CAD model, geometric features or learned keypoints. The initial result is an object pose relative to the camera.
Symmetry
Round or repetitive shapes may have several technically equivalent orientations. The process step determines which differences actually matter.
Quality threshold
A detected pose is used only when visible geometry, fitting error and measurement confidence remain within predefined limits.
Coordinates
From camera frame to robot base
With a fixed camera, model matching produces the object pose in camera coordinates. A calibrated transformation describes the camera's position relative to the robot base. Combining both transformations in the correct order gives the object pose in robot coordinates.
With an eye-in-hand camera, the robot's current arm pose also enters the chain. A reversed transformation, outdated calibration or shifted camera mount displaces every robot target, even when model matching itself appears stable.
Read more about establishing and checking this relationship in hand-eye calibration.
Transformation chain
Camera → object:
from 3D measurement and model matching
Robot base → camera:
from calibration
Robot base → object:
the target for the next robot step
Error propagation
A detected part is not yet a usable robot target
Measurement noise, surfaces hidden by other parts, model differences and calibration error all propagate to the final target. In particular, a small orientation error can cause a noticeable position error farther away from the estimated centre.
Define an error budget for the complete chain. Permitted uncertainty follows from the gripper, fit, insertion clearance or inspection limit, not from one camera specification in isolation.
Production value
Expose uncertainty before the robot moves
Less downtime
Rejecting an uncertain target before motion prevents collisions, missed picks and jammed insertions.
Focused diagnosis
Separate measurement quality, model fitting and calibration so maintenance does not troubleshoot by guesswork.
Manageable capacity
Measure how often a pose is accepted, captured again or requires operator assistance.
After acceptance, grasp planning uses the object pose to evaluate reachable and stable contact points.
Next step
Test the complete measurement and calibration chain
Share part geometry, camera arrangement, robot configuration and process tolerance. A representative trial can show which poses are reliable enough for the intended robot task.