Adaptive neural network control of tendon-driven mechanisms with elastic tendons

Nonlinear Robust Control of Tendon–Driven Robot Manipulators

“…Among other remote power transmitting methods, tendon-driven transmission systems present a less noisy, clean (as they do not require lubrication), and shock absorbent characteristics. Therefore tendondriven mechanisms have been used in the design of both small [1][2][3] and large robot manipulators [4][5][6]. However the use of tendon-driven actuation has been more popular in dexterous hands [2,3,7] as the resultant task-space motion in robotic hand designs does not need to be accurate and relatively simple controllers can handle desired objectives.…”

“…However, when the dynamics of the power transmission system is added to the system model, the control design problem becomes more complicated mostly due to the addition of extra dynamics and therefore additional uncertainties. On top of this, for an acceptable tracking performance, tendon elasticity in the transmission system, specific to tendon-driven systems, has to be considered in the controller design [5,8].…”

“…To name some; the development and performance of an inner-loop adaptive and robust outer-loop controller was investigated in [11], while in [12], robotic mechanisms with redundant tendons were investigated and in [13], authors propose a puller-follower controller both compliant and non-compliant antagonistic tendon drives in robotic systems. In [5], Kobayashi and Ozawa presented an adaptive and an adaptive neural network based controller for tendon-driven robotic mechanisms with elastic tendons. In [14], Nakayama and Fujimoto tackled the tracking control of tendondriven robots by applying the delayed reflexive force feedback.…”

“…In [17], Wimbock et al proposed an application of the Immersion and Invariance type framework to tendondriven systems with variable stiffness. Among the above cited work, the only work that considered the uncertainties in the system dynamics was given in [5], however the proposed adaptive controller required the measurement of the second and third time derivatives of link position measurements (see assumption 2 of [5]) which are usually not available.…”

“…Review of the past research on control of tendondriven robot manipulators has revealed that almost all of the control designs required accurate knowledge of the system model along with exact knowledge of model parameters, and according to the authors' best knowledge, the only work considered parametric uncertainties was [5] which utilized link acceleration, and jerk measurements. In this study, we aim to design a robust controller that does not require acceleration and jerk measurements.…”