Decoupled torque control of tendon-driven fingers with tension management

Abstract

To facilitate human assembly tasks, Robonaut 2 is equipped with a dexterous, compact hand featuring fingers driven remotely by tendons. This work outlines the force-control strategy for the fingers, which are actuated by an "n +1" tendon arrangement. Existing tendon-driven fingers have applied force control through independent tension controllers on each tendon, in other words, in the tendon space. The coupled kinematics of the tendons, however, cause such controllers to exhibit a transient coupling in their response. This problem can be resolved by alternatively framing the controllers in the joint space of the finger. A joint-space torque control law is proposed here that demonstrates a decoupled response with a faster settling time than an equivalent tendon-space formulation. In addition, a tension distribution algorithm is presented here to translate joint torque commands into tendon tensions. It guarantees that each tendon tension respects both an upper and a lower bound, using an efficient, finitely-convergent algorithm. These two contributions provide for a compliant, well-controlled hand, aptly suited for unstructured interaction.

Analysis of humanoid robotics for nuclear disaster management incorporated with biomechanics

“…Fig ure 4 shows the sim pli fied fin ger me chan ics of the hu man oid's hand. Abdallah et al [12] stud ied the me chan ics of the hand of hu man oids. Let us think the con fig u ra tions as k, t, d, and f are the col umn ma trices of joint an gles, ac tu ated joint torques, col umn matri ces of ten don po si tions, and ten sions, re spec tively [12].…”

“…Abdallah et al [12] stud ied the me chan ics of the hand of hu man oids. Let us think the con fig u ra tions as k, t, d, and f are the col umn ma trices of joint an gles, ac tu ated joint torques, col umn matri ces of ten don po si tions, and ten sions, re spec tively [12]. So, t = Rf (4) where R has the el e ments of the signed pul ley ra dii in the ten don rout ing path [13].…”

Adaptive neural network Dynamic Surface Control: An evaluation on the musculoskeletal robot Anthrob

“…Most tendon-driven controllers use feedback linearization by linearizing (6) and introducing some linear control law [7], [13]. Hence, a tendon force f d is computed for a given movement and subsequently controlled by a low-level force controller.…”

“…The angle error ∆q according to (13) was utilized with the positive definite gains h 1 and h 2 to estimate the system states. Since the controller utilizes the same model of the system dynamics to compensate for any non-linearities, the result of γ 0 is essentially equal to the reference accelerationq d .…”

Decoupled Control of Position and / or Force of Tendon Driven Fingers

“…However, such arrangements cause couplings, because tendons to distant joints pass by the more proximal joints. In this way it is not trivial to control the joints in a decoupled way [11], [12], [13], [14], [15], [16]. In particular it is shown in this paper that existing approaches for decoupled joint torque control, e.g.…”