Robot hands from joint count to a reliable grasp

Joint motion and independent actuation are different counts

Shadow's classic Dexterous Hand gives a concrete example. Its published comparison lists 20 actuators and 24 joints, including under-actuated movements. The same table distinguishes wrist joints and installed tactile fingertips. The Lite variants have different configurations. [1]

A mechanically coupled movement can follow another joint rather than receive an independent command. Therefore, avoid comparing one manufacturer's motor count with another manufacturer's total joint count as though they were the same quantity.

For a proposed task, ask which contacts must be controlled independently. Picking a rigid block between two faces places different demands on the hand than turning a key while maintaining its position in a lock.

Scroll sideways for all columns.

Compare a dexterous hand with a task-specific gripper

Robotiq's adaptive grippers provide a two-finger comparison, while Shadow supplies multi-finger research hands. These products address different combinations of contact geometry, control and integration. Product documentation should be used to identify the exact opening, payload conditions and interfaces. [1] [2]

In an illustrative box-transfer job, a simple opposing grasp may complete the requirement. A multi-finger hand becomes relevant when the job calls for additional contact choices or movement within the grasp. Compare completed work and recovery requirements before treating finger count as the objective.

Know what each sensor actually measures

Shadow documents tendon load sensors and tactile fingertip options as separate elements. That distinction is useful when reading a hand specification. A measurement in the drive system and a measurement at the contact surface are taken in different places. [1]

For a research trial, retain contact measurements alongside motor commands and images. A rising motor command can have several explanations, including contact or a mechanical obstruction. The trial should establish how the system distinguishes them.

Specify where a slip is detected and what action follows. More sensing points do not automatically provide a tested slip-recovery policy. The policy and its trial results need their own evidence.

A holding-force calculation needs explicit assumptions

Consider an idealized two-finger pinch holding a 0.5 kg object vertically. Assume equal normal force at both fingers, a friction coefficient of 0.5, no acceleration and no other support. The object's weight is about 4.91 N. In this simplified model, each finger needs at least 4.91 N of normal force because the two friction forces sum to the object's weight.

This is a teaching calculation, not a safe operating limit. Real contact area, surface contamination, acceleration and deformation can invalidate the assumptions. A product's quoted grip force may also use a different definition. Compare definitions before placing its number into the equation.

Test pickup, retention and release separately

An illustrative 30-trial test might record 27 pickups, 24 successful carries and 21 correct releases into a fixture. The full-task completion rate is 70%, while pickup alone is 90%. Separate those stages to locate the failure.

For in-hand manipulation, additionally record whether the object changed orientation without being set down or supported externally. Identify the starting grasp and permitted assistance. Include wear, contact-surface replacement and calibration in a repeated-use evaluation.

Keep hand research separate from prosthetic fitting. A robotic component attached to an arm and a device worn by a person have different interfaces, constraints and evidence requirements.