Stability through design : towards dexterous in-hand manipulation in tendon-driven robotic hands
While dexterous robotic manipulation research has made significant advancements in the last two decades in areas of sensors, control strategies, perception and planning, the abilities of robotic hands in unstructured and unpredictable environments are limited. Specifically, a few researchers have shown promising manipulation results with stiffness controllers, which allow for the generation of fingertip forces as a function of displacement. In terms of mechanical design, robotic hands have been converging towards low-inertia, passively compliant, tendon-driven strategies for agility and robustness against environmental impacts. As tendon driven robotic fingers are serial chain systems, various routing strategies with passively compliant tendons lead to unique multi-articular stiffness coupling between the degrees of freedom. The performance of manipulation controllers is highly dependent on the passive properties of the fingers. While tendon-driven robotic fingers are widely accepted to be advantageous for manipulation and stiffness controllers have shown promising results, currently there is no methodology for a thorough analysis of the effects of various compliance arrangements in the fingers on the closed loop properties of stiffness controllers. As a result, we don't have the ability to reliably predict the boundaries of stable operation and determine the limitations due to mechanical parameters for tendon-driven robotic fingers. We also don't have a quantifiable way of exploiting the design of mechanical elements to intrinsically improve the dexterity and robustness of robotic in-hand manipulation by augmenting the controllers.
In this dissertation, I present a systematic methodology for analyzing the effects arrangements of passive compliance on tendon driven robotic joints implementing stiffness control strategies. To begin, I develop generalizable comprehensive mechanical models of compliant tendon driven robotic fingers. Then, I identify the various arrangements of passive stiffness elements and analyze their role in the performance of various stiffness control strategies and subsequently towards dexterity.
I have analyzed the achievable joint stiffness control boundaries of tendon-driven robotic fingers implementing joint stiffness control leading to a first of its kind generalizable stability boundary that can be applied to robotic fingers with any degrees of freedom and tendon routing strategy. Then, I extend the analysis to Cartesian fingertip space as object manipulation requires accurate control of fingertip force directions and magnitudes. I use the analysis to identify all the mechanical design features and dynamic parameters that have a direct impact on controller stability. I have isolated compliance in parallel to actuators as a significant element for optimization. Optimal linear and nonlinear parallel compliance found using the analysis improved the stability and force tracking accuracy of Cartesian stiffness control even in the presence of external forces. Such features are ideal for in-hand manipulation. Finally, I extend the stability analysis to object-space stiffness controller and optimize linear and nonlinear parallel compliance for improved dexterity, accuracy and robustness of in-hand manipulation.
My research not only allows for an accurate prediction of the behavior of stiffness controlled tendon-driven robotic hands but also leads to a mechanical design paradigm informed by the stability of robotic hands allowing for the design of intrinsically stable, robust and dexterous robotic hands that take us one step closer to human-like dexterity.