The 3D Adaptive Compliant Gripper compares one preset closing force against a slip-triggered adaptive one. This 2D companion opens the gripper's own mechanism up: each of the three fingers is an underactuated chain — one tendon per finger, two joints, no per-joint motor. Every joint is a torsional spring:
joint torque: I·θ̈ = τ_tendon − k·θ − c·θ̇ + τ_contact
contact→joint: τ_contact = Σ F_contact · (P_contact − P_pivot)⊥ (virtual-work / Jacobian-transpose coupling)
- τ_tendon is identical for every joint on a finger — one pull, many joints. That's underactuation: 3 tendons drive 6 joints.
- Before any contact, the softer-feeling joint (lower effective resistance) closes first, exactly like a real underactuated finger.
- The instant a link touches the object, the contact's reaction torque at that joint rises to balance the tendon torque there — that joint stops advancing while the same tendon keeps closing the other joint, so the finger keeps wrapping past the contact point.
- No sensor ever measures the object's shape — the differing torques at each spring joint are the only thing doing the adapting. That's why a circle, a square and an irregular blob each end up held by a different, unprogrammed finger posture.
Simplifications: a planar rigid-link model with a penalty (spring) contact and no friction, so it shows how the fingers wrap a shape but not slip or the friction cone; the tendon is an ideal, frictionless torque source shared by both joints and the object is fixed at the hub centre (it is not pushed around). The picture is Blender-rendered art over live physics.
Independent checks: running...
The lower chart sums the penetration-based contact force per finger in real time — the force-distribution readout is a side-effect of the mechanics, never an input to them.