As intelligent robots become more pervasive in the facilitation and execution of historically human-centric tasks, we, as roboticists, must continue to improve upon our methodologies for intuitive and efficient human-robot interaction (HRI). This is especially relevant to robotic systems that are intended to operate in close proximity to humans where physical contact is either intentional or inevitable. The vast majority of research in physical human-robot interaction (pHRI) has focused on exclusively anthropomorphic realizations of interactions that involve complex and rigid traditional serial robotic systems, which rely on sophisticated sensing and control implementations to accommodate physical contact. Furthermore, the breadth of the pHRI research domain has remained limited by the largely anthropocentric perspective of prioritizing human-like interactions as well as the overwhelming emphasis placed on contact avoidance as an essential feature in autonomous and mobile robotic systems. In contrast, the somewhat recent proliferation of soft robotic systems, like tensegrity robots, built with an intrinsic tolerance for physical contact have shown tremendous promise as platforms for enabling pHRI.

Tensegrity robots are a class of soft robotic systems whose structures consist of a set of rigid bodies suspended in isolation via a network of cable elements. The advantages of a tensegrity robot include low density, configurable compliance, and structural resilience at the cost of greater complexity in modeling and control. These unique mechanical characteristics make tensegrity robots well-suited to applications that demand robustness to physical contact. In this dissertation, we examine the design and implementation of a force-sensing tensegrity as a robotic platform for enabling novel physical interactions and for exploring new avenues for pHRI with compliant robotic systems. First, we explore the potential for a new language of pHRI that leverages non-anthropomorphic, compliant, and mobile robotic systems. We then present the Class-1 spherical six-bar tensegrity topology as a scaffolding for implementing the detection of physical human-robot interactions. Several force-sensing tensegrity prototypes are designed, constructed, and tested to explore the capacity for reliable contact detection. To demonstrate the ability of the force-sensing tensegrity to distinguish between physical interactions, we propose a methodology for inferring intent from physical interactions using a supervised learning framework that features contemporary classification algorithms including deep neural networks. Additionally, we conduct a series of human subject experiments to examine the intuitiveness of physical interaction with the tensegrity as well as the robustness and generalizability of the aforementioned supervised learning framework.

There are broad implications from these results on the future of pHRI research leveraging similar robotic implementations, which could be capable of offering completely new functionalities for physical interaction. The methodologies and frameworks presented here can be extended to various tensegrity topologies, fully actuated tensegrity platforms, and even compliant robotic systems outside the domain of tensegrities. As a result, we hope that pHRI researchers will be inspired to utilize compliant systems like our force-sensing tensegrity as viable platforms for investigating physical interaction. In summation, the problems addressed here constitute an exciting and potentially paradigm-shifting investigation of the utility of tensegrity robots as platforms for a new language and embodiment of pHRI with compliant robotic systems.