Knowledge · Robotics

Grasp planning: from object pose to an executable grasp

Once a part's position and orientation are known, grasp planning determines where the gripper can make contact and whether the robot can execute that grasp safely. Not every geometrically possible grasp is reachable, stable or collision-free.

Industrial 3D camera for robot guidance

Decision chain

Generate, reject and rank candidates

A grasp candidate describes the gripper's position and orientation relative to the part. Software generates candidates on suitable faces, edges or predefined contact points. It then retains only those that satisfy the defined process constraints.

Stability

Jaw opening, contact area, suction cup diameter and centre of gravity determine whether the part remains controlled while it is lifted and moved.

Measurement confidence

Surfaces hidden by neighbouring parts were not measured directly. A candidate that relies heavily on this hidden geometry should rank lower or be rejected.

Reachability

The robot must reach both the pre-grasp and grasp poses without exceeding its joint limits.

Clearance

Models of the gripper, fingers, wrist and part are checked against the bin wall, neighbouring parts and fixtures during approach and withdrawal.

Production example

Steel brackets in a deep bin

Pose estimation provides several possible contact points for one visible bracket. A flange grasp is rejected because a neighbouring part occupies the space required by the fingers. A side grasp is rejected because the robot cannot achieve a valid wrist pose near the bin wall. A third candidate has clear contact faces and a collision-free approach, so it is selected.

After closing, the cell uses jaw position or vacuum pressure to verify that exactly one part was picked. Without confirmation, the controller tries another candidate or requests a new capture.

Production value

Evaluate more than a successful demonstration

First attempt

Useful ranking reduces missed picks, extra captures and recovery moves.

Cycle time

Candidate generation, collision checks and alternatives must fit the available process time.

Intervention

Define when the cell retries autonomously and when an operator is required.

Test representative part variants, fill levels and unfavourable orientations. Relate the results to the cycle time of the complete application, not only the planner's computation time.

Next step

Test parts, gripper and bin as one system

Share part geometry, the gripper model, bin dimensions and process limits. A trial can then show which grasps are measurable, reachable and verifiable.