Control of muscle strain energy as a robust means to produce slow and accurate finger movements:
Proof of concept via hardware and cadaver implementation
Heiko Hoffmann, Jason Kutch, Manish Kurse, Francisco Valero-Cuevas
Biomedical Engineering and Biokinesiology and Physical Therapy, University of Southern California

The neural control of slow and accurate finger movements remains an open question. A possible reason for this challenge is the critical intricate dynamical balance across muscles in this context. That is, given the coupled kinematics of the multi-articular tendons actuating the fingers, even small errors in musculotendon forces or excursions can cause undesirable joint kinematics, sudden accelerations, or overloading of the tendons and joints. While the available theories of motor control may well apply to this problem, to our knowledge they have not been tested. As a means to test competing hypotheses of motor control in this context, we created a computer–controlled tendon actuation system capable of driving both tendon-driven mechanical fingers and human cadaver fingers. Cadaveric studies are at the origins of anatomical science; only recently, however, have finger tendons been controlled to estimate, for example, moment arms [1] or static force production [2]. Tendon-driven robotic systems have also been proposed to study finger function [3]. Our initial experiments centered on testing the feasibility of position control, force control, and impedance control to produce slow and accurate motions in a mechanical finger and two cadaveric index fingers (male, left hand).

We found that both position and force control strategies in isolation were impractical. Position control required accurate control of tendon excursions to match the kinematic constraints; small errors lead to either slack or overloaded tendons. Force control resulted in unstable finger positions. In the mechanical finger, we could not stabilize desired postures. In the cadaveric fingers, a given posture under force control was not retained after perturbing either force or posture. In contrast, programming the tendon actuators to behave like tunable springs lead to robust production of slow and accurate finger movements in both mechanical and cadaveric fingers. In the cadaver fingers, a given stabilized posture was retained after manually perturbing this posture. Moreover, we were able to produce slow tapping movements (30 s per tap). The programmed springs' individual strain energies were regulated by setting their fictitious resting lengths (spring stiffness: 3 N/mm, initial strain: 1 mm).

Our results suggest that tunable springs are used for low-level muscle control to enable both replication and learning of slow and accurate finger motions. To our knowledge, this is the first time that all tendons of cadaveric fingers have been used to control slow and accurate movements. This paradigm allows us to compare and contrast alternative neural controllers for slow and accurate finger movements, while fully considering the anatomical complexity of fingers, the nonlinearities of their tendinous networks, their low mass, and soft tissue deformations. Understanding the necessary low-level control properties of muscles in this context will promote understanding the neural control of manipulation and develop robust control strategies for robotic and prosthetic limbs and electrically-stimulated paralyzed limbs. This work has been funded in part by grants NIH R01 050520 and NIH R01 052345 to FVC.

[1] An KN, Ueba Y, Chao WP, Cooney EY, and Linscheid RL. J Biomech 16: 419-425, 1983.
[2] Valero-Cuevas FJ, Towles JD, and Hentz VR. J Biomech 33: 1601-1609, 2000.
[3] Bidic S, Imbriglia J, and Matsuoka Y. Proc Annual Meeting of the American Society for Surgery of the Hand, 2003.

Back to Publications