Manipulation

Manipulation is where Parts I–III stop being chapters and start being a shift. You need a pose of an object (vision, features, maybe a net), a grasp that statics would bless, an arm trajectory that does not hit the table, and a release that does not topple the mug. Any one of those can be a paper. The teaching job is to keep the pipeline visible so students do not spend six weeks on a color threshold and call it manipulation.

Inverse kinematics returns for real: the grasp frame in the world must become joint angles, then a motion that respects limits and collisions. Jacobian-transpose control is a gentle way to press, insert, or follow a surface when you would rather command a force than a brittle pose. If the previous lessons felt abstract, this is their court date.

Pick-and-place is not “IK then open.” The object may move. The grasp you planned in the camera frame is stale after you reach. A human hand is in the workspace. You need the task-execution lesson: modes, timeouts, a reflex if contact comes early. Motion planning here is often in configuration space with a swept volume of the hand plus object — after the grasp, the robot changed shape.

Compliance is how you survive being wrong. A rigid position command into a slightly misplaced hole will either miss or bend something. A bit of give — a spring, a force loop, a guarded move along the table — turns geometry errors into slides. Industry learned this before academia was polite about it. Teach one guarded move: move down until force, then slide until a second force.

If the lab has no arm, mime the pipeline with a webcam and a human hand as the actuator, or use the mobile base to “place” a block into a taped square. The ideas travel. Then read the book chapter for the formal treatment this companion refuses to copy.