work led to important medical robot commercialization efforts, such as Hansen Medical’s tendon-actuated cardiac catheter. In addition, a tendon-actuated design was proposed in which the flexible backbone was replaced by a series of short cylindrical links connected by spherical joints. This design became the basis of a surgical robot commercialized by Medrobotics. During the 2000s, the concept of concentric tube robots was first introduced, but a more complete description of the design principles and kinematic model for this architecture was finished only in 2010 (62). During the decade 2010–2020, continuum robot research was focused in four areas: (i) incorporating external contacts and loads in robot modeling and control, (ii) developing methods to control robot stiffness, (iii) creating “soft” continuum robots, and (iv) the design of continuum robots for specific clinical applications. Each is described below. Extending kinematic models to consider external contacts and loads In many medical applications, a robot will contact tissue not only at its tip but also at many locations along its length. Unlike rigid robots, these contact forces can produce appreciable deformation of a continuum robot, leading to large errors in the kinematic map relating, e.g., tendon tension to tip position and orientation. An important research thrust has been to include external loading in the kinematic model (63) and to infer external loads from the kinematic input variables, e.g., tendon tension forces (64). Alternatively, a model-free approach has been proposed in which the contactconstrained kinematic model is estimated during task execution (65). For model-based control methods, an alternative to inferring external loads from kinematic inputs is to directly sense them. Although the creation of a distributed sensing skin at the size scale and price point appropriate for medical interventions remains an open problem, a noteworthy effort over the decade has gone into the development of sensors that can estimate robot shape (66). Stiffness control In contrast to rigid robots, the inherent flexibility of continuum robots enhances their safety during navigation through the body to a surgical site. Surgical tasks, however, involve applying forces to tissue, and the lower tip stiffness of continuum robots requires larger robot displacements to produce a given force. The task-based force level together with limited volume available to maneuver the robot defines a minimum tip stiffness needed to perform the task. Important work over the decade has developed mechanical design methods for enhancing and controlling continuum robot stiffness, e.g., by incorporating layer jamming in the flexural components (67). For those situations when the inherent stiffness is sufficient, control algorithms have been developed that modify the kinematic inputs to achieve a desired tip stiffness (68). Downloaded from https://www.science.org on November 16, 2021 Dupont et al., Sci. Robot. 6, eabi8017 (2021) 10 November 2021 SCIENCE ROBOTICS | REVIEW 11 of 15 Soft continuum robots Continuum robots are often fabricated from compliant polymeric materials, and some of the earliest examples were actuated pneumatically or hydraulically—the two features typically used to define “soft” robots. With a few exceptions, however, medical continuum robots have eschewed gas or fluid actuation, which tends to increase modeling complexity and response time. With the explosive growth in soft robotics over the past decade, however, these actuation methods and the use of even more compliant materials are now being explored for medical applications (69). Application-specific continuum robot design In addition to deepening the technological toolbox, researchers have also collaborated with clinicians to create robotic systems designed to perform specific procedures. For example, Ding et al. (16) produced a single-port system for abdominal surgery. Once inserted into the abdomen, two multibackbone continuum arms along with a conventionally jointed stereoendoscopic arm extend from a single sheath to create an anthropomorphic representation of the surgeon’s head and arms. This technology was licensed for commercialization by Titan Medical. As a second example, the system of (20) explores the use of two concentric tube robots together with a separate passive endoscope for transnasal skull base surgery. This system was an important early demonstration of how the concentric tube architecture, along with the theoretical modeling developed to support it, could provide the workspace, stiffness, and manipulability necessary to perform actual neurosurgical tasks. The past decade has provided a maturation of the fundamental techniques for designing and modeling the various continuum robot architectures. Although this research is largely complete, the availability of new sensing technologies will likely spur the development of improved sensor-based control techniques. For example, fiber Bragg grating sensors, a very expensive technology, is the main shape-sensing modality that has been investigated (66). An inexpensive alternative technology would likely result in a new generation of control algorithms. Furthermore, we will likely see continued interest in applying soft robotics to produce alternative robot designs and learning/artificial intelligence applied to robot