Flexible Robotics in Medicine: A Design Journey of Motion Generation Mechanisms and Biorobotic System Development

Preface: A design journey of biorobotic motion generation mechanisms and flexible continuum system development

The primary motivation of the book is to document our design journey of biorobotic motion generation mechanisms and flexible continuum system development. Practical hands-on trials and errors for the growing clinical needs concurrently promote critical thinking of flexible robotics in terms of platform creation. The book involves problem-solving and design-centric approaches toward research on flexible robotic systems. The robotic development pipeline includes the state of the arts, the assessments of the clinical background and unmet needs, the engineering principles involved, model analysis, design evaluations, characterization, and modeling of biomedical engineering systems. The book is primarily for biorobotics beginners to have an overview of the design journey in continuum motion generation mechanisms and flexible surgical robotics.

The book focuses on the motion generation methods and proof-of-concept prototyping involved in the journey of design-centric biomedical robotics. The problems are generally clinically driven in real-world scenarios, so the solution domain is typically open and complex. For example, a common clinical problem in the subsequent chapters is to design mini-endoscopic flexible mechanisms that can navigate in a nasal cavity for surveillance.

The journey of design-centric projects in developing flexible mechanisms typically starts from fundamental conceptualizations, going through the development pipeline until the final prototypes. Design-centric biorobotic prototyping being a highly iterative process, the generations of designs of each chapter show logical flow and justifications to overcoming problems and limitations and ultimately to bring a design concept to an optimal prototype reality. Literature critique and technology comparisons are to expose the idea of innovation. The first-hand interactions and review in clinic real-world problems reduce abstractions of the design thinking and add a unique dimension to each chapter of biomedical robotic design.

The chapters of the book typically cover the critical milestone in developing a prototype, including clinical background, needs, design conceptualization, specifications, competitiveness survey, design rationale, modeling, verification, patent review, and regulatory affairs.

The design concepts of most chapters have an emphasis on the clinical background and needs, which subsequently lead ideas into real solutions recognized by the community. With prototype-specific regulatory requirements in mind, we identify and evaluate a prototype classification and justify the unique contributions in comparison with the competing technologies.

User specifications emphasize the needs of the design project and the process of parameterizing those needs, in contrast with similar states of the art. Design rationale and verification discuss the experiments and benchmarking tests that apply to the prototype claims and design functionality. The design-centric chapters also identify engineering testing standards and how a measure applies to design. An example of standards to reference is the ASTM standards.

Most chapters illustrate the process of reaching the final prototype demonstrations (as in the supplementary videos), and it was fascinating to observe the creative side of our members addressing the same problem with different approaches. Remarkable prototypes further push forward to clinical cadaveric study to validate the design concepts. 3D printing technology is typically for the rapid fabrication of manipulator segments and proof-of-concept study in the chapters. Various design iterations with 3D printing allowed the rapid progression through the design process of a series of flexible manipulators. The fabrication of small components would have been difficult with the current fabrication capabilities of a 3D printer, and an alternative way will be upscaled prototyping for a proof-of-concept study.

A competitiveness analysis component is to understand the states of the arts and a brief survey of bedside potential. Patents are integral to intellectual property protection, and the design-centric chapters documented the process of defining features and how the design would break into claims in a patent. The patent review encompasses the novel aspects of the design and contrasts their composition with other patents. Most design-centric chapters analyze from a biomedical engineering standpoint, which involves modeling and verifications during the device developments. The combination of these elements strengthens and promotes future cutting-edge designs.

Through the documentation of each chapter’s design journey, we can observe the growth of the design through the various stages, where the final design encompasses parts of creativity from the different stages. The book is mostly the prototype developments from our design-centric projects and modules at the Department of Biomedical Engineering at the National University of Singapore. I highly appreciate the investigation efforts from my collaborators and my group members at different levels ranging from undergraduates, graduates, to postdoc fellows. We envision the book primarily for biorobotics beginners to have a glimpse of the design journey in evolving various motion generation mechanisms for medical applications.

Chapter 1

Slender snake-like endoscopic robots in surgery

Shumei Yu¹,², Wenjun Xu²,³ and Hongliang Ren²,    ¹School of Mechanical and Electrical Engineering, Soochow University, P.R. China,    ²Department of Biomedical Engineering, National University of Singapore, Singapore,    ³Peng Cheng Laboratory, Shenzhen, P.R. China

Abstract

For surgery, the targets inside the human body have complicated anatomical structures, which make access to the operating region challenging with high risks of bleeding and infections. Though keyhole surgical robots with articulated arms help surgeons to overcome the access difficulties in part, narrow and curved cavities require robots to be more compact and flexible. Thus snake-like robots are developed to cater to the demand because of their slim body and hyperredundancy in movement. This survey presents state-of-the-art knowledge about snake-like robots for robotic surgeries. We have investigated the commercial products, representative research prototypes, and new mechanical designs of the surgery-oriented snake-like robots. Additionally, for safe and precise human–robot interaction, modeling, sensing, and control of snake-like robots have been summarized.

Keywords

Snake-like robots; surgical robots; kinematic modeling; statics; dynamics; compensation; force sensing; shape sensing; workspace; motion planning; control

For surgery, the targets inside the human body have complicated anatomical structures, which make access to the operating region challenging with high risks of bleeding and infections. Though keyhole surgical robots with articulated arms help surgeons to overcome the access difficulties in part, narrow and curved cavities require robots to be more compact and flexible. Thus snake-like robots are developed to cater to the demand because of their slim body and hyper-redundancy in movement. This survey presents state-of-the-art knowledge about snake-like robots for robotic surgeries. We have investigated the commercial products, representative research prototypes, and new mechanical designs of the surgery-oriented snake-like robots. Additionally, for safe and precise human-robot interaction, modeling, sensing, and control of snake-like robots have been summarized.

1.1 Introduction

As engineering and medical science developed, robotics in the surgical field has made significant progress in the aspects of accuracy, efficiency, and safety. Among the boom of surgical robots market, the successful units in corresponding surgical areas include Da Vinci in laparoscopic surgery, CyberKnife in radiosurgery, MAKOplasty in a knee or hip replacement, Renaissance in spine surgery. These units break through the traditional operations by minimal invasiveness, clearer field of view, accurate targeting, navigation, and manipulation. Nevertheless, for the complex anatomical structures with narrow and curly cavities, such as a paranasal sinus, nasal cavity, antrum auris, pharynx, and larynx, it is difficult to reach and operate by traditional devices. Although endoscopic technology has been developed, surgeons have to deflect their attention partially to manipulations from inspections. The migration of industrial robots is facing challenges with narrow and curly cavity environments. Thus robots with the ability of flexible motion bring about the interests of the researchers and developers.

Snake, which goes through 100 million years of evolution, lives in nature almost around the world with high adaptability, still keeps the long, slim, and limbless body feature. Snake locomotion has attracted the interests of scientists in bionics and robotics for a long time. Since Shigeo Hirose in Tokyo Institute of Technology developed the first snake-like robot in the 1970s, studies on snake-like robots’ locomotion mechanism, mechanical design, gait generation, and motion control have been continuously published. Among the motion generation theories, gaits generated based on the curves are the most popular, exceeding the central pattern generator (CPG) and dynamic model–based methods. Chirikjian developed the backbone curve theory, and Choset studied the discretization of a curve in snake-like robot motion generations. Due to the hyperredundancy in DOF, snake-like robots can be used as a flexible manipulator to inspect spaces that are difficult to reach, for example, products by OC Robotics as inspectors. Inspired by the industrial usages, researchers have been trying to study snake-like robots that can be used for operational purposes.

Traditional snake-like robots have articulated rigid links. Due to their bulky shape, although they are found in certain studies that cater to stomach and intestine biopsy applications [1], they are of low priority in consideration of surgical applications. Continuum robots that take the form of cable-driven, concentric tube, catheter, steering needle fit the operation environment better than the snake-like robots of traditional modality. However, continuum robots in a surgical area have a long and slim shape and move like a snake, so they are often named as snake-like robots likewise. As has been surveyed previously [2], snake-like robots have been developed for applications in neurosurgery, otolaryngology, cardiac surgery, vascular surgery, abdominal surgery, and urological surgery. Take the cable-driven snake-like robot; for example, the actuation mechanism introduces backlash in the movement [3]. Additionally, the friction force is hard to be recognized to get an accurate hysteresis model. Besides, when the robots move inside the human body, it is challenging to build the interaction model. The sensing of the position, configuration, and force are difficult issues due to their downscaled size. Therefore accurate control for this type of robot to reach the target and accomplish the operations such as inspection, biopsy, cutting, and suturing is difficult. This survey discusses snake-like robots in surgical applications and summarizes the recent progress in mechanical design, modeling, sensing, and control. Among the contents, the authors will emphasize the motion compensation techniques, workspace analysis, motion planning, and control of the robots, which have not been surveyed intensively in previous studies.

1.2 Snake-like robots for surgery

Because of the similarity on shapes, endoscopes can be reckoned as the predecessor of the snake-like robot for surgical usage. Equipped with cameras and essential transmission mechanisms, an endoscope can be used to inspect or do biopsy tasks inside cavities of the human body. However, surgeons found the functions of the existing endoscopes challenging to satisfy the need for more complicated manipulations. Furthermore, an endoscope is at high risk of causing perforation, indigitation in clinical usage. To cater to the demand of intra-lumen exploration, researchers have made their efforts on various types of actuation and transmission designs to develop compact snake-like robots in the past decades. Most of them take the form of cable-driven, discrete joints, concentric tubes, continuum joints, and articulated rigid-link structures for surgical applications.

1.2.1 Commercial products

In this subsection, we summarize the commercialized snake-like robots for surgery. Table 1.1 shows the current surgical systems, their developers, and their application areas. Because of the massive success of Da Vinci in minimally invasive surgery (MIS), most of the snake-like surgical robots developers turned their attention on natural orifice transluminal endoscopic surgery (NOTES), for example, ViaCath [6] and Master [7]. For a subclass of NOTES such as colonoscopy and transoral surgery, a specialized product for a specific natural orifice such as NeoGuide [4] and Flex [5] emerged. CardioARM [8] is intended for specific cardiac surgery.

Table 1.1

Most of the commercialized snake-like surgical robots adopted the cable-driven structure as driving mechanisms. The NeoGuide has 16 segments, each of which with the 8 cm length can be controlled in the desired direction. The ViaCath and Master were designed for the single-port laparoscopic surgery as a subclass of MIS. As required by the intended tasks, the robots have a long endoscope inserted through a dedicated lumen. Flexible instruments can be hidden in the endoscope tubular trocar first and then deployed after the system has been in the targeted areas. Fig. 1.1A shows the mostly used cable-driven type taking the one primary backbone and two cables for an example. The Flex and CardioARM take the design of the inner tube and outer tube structure, which are wire-driven alternatively, forming the follow-the-leader motion control in the operation area. The inner and outer tube structure is shown in Fig. 1.1B.

Figure 1.1 Typical mechanical structure of snake-like surgical robots. (A) Cable-driven mechanism with primary backbone and two cables. (B) Front and side view of the cross inner and outer tube snake surgical robot.

1.2.2 Typical mechanical design

Flexible snake-like surgical robots can be realized by the tendon-driven (or can be called cable-driven, wire-driven) structures.

1.2.2.1 Tendon-driven mechanisms for surgical robots

MIS or NOTES require a continuum manipulator with a small diameter but lumen for the observation and end-effector instruments. Except for the compact design of the shape, it should be convenient to be sterilized. Furthermore, except for the electrosurgical unit, the inside manipulator should not release heat due to motor drive. Considering all the above factors, the tendon-driven mechanism fits the tasks mostly.

Table 1.2 summarized the typical tendon-driven snake-like surgical robots since 2011. The mechanism designs can be classified into four groups, namely the one primary backbone, four primary backbones, no primary backbone, and concentric tubes. For the one primary backbone, the first subgroup used disks as tendon holders, which has been shown in Figs. 1.1A and 1.2B. Following are examples: Bajo et al. [12] separately built the tendon-driven continuum robot using concentric disks connected in series with one primary backbone; Sen et al. [13] and Ouyang et al. [18] introduced the three-segment continuum cable-driven robot for a larger workspace. The second subgroup made various evolutions to the tendon holders, such as Li and Du [19] used articulated spherical joints to connect nodes as the holder for four secondary backbones in a single segment of the robot whose sketch can be seen in Fig. 1.2A; Kato et al. [16] designed a hinged wire guide to obtain the bending room and expanded the degree of freedom (DOF) of their robot by connecting two sections of the tendon-driven mechanism. The third subgroup evolved the distribution of the driving tendons. Gao et al. [17] developed the cross-helical tendon-driven manipulator, which can be bent into S shape with only one segment, and Fig. 1.2B shows the abbreviated mechanism. Notched cylinder with holes for driving-wires was designed as a continuum robot to generate flexible bending [11]. For robots with no primary backbone, Moses et al. [11] proposed a high-strength cable-driving dexterous manipulator with the lumen provided by a rigid guide cannula, as is shown in Fig. 1.2C. Kim et al. [15] realized neutral line and stiffness control by designing the manipulator as asymmetric wire-connected rolling joints. As for the robot of concentric tubes, a robot with four primary backbones is the i² Snake newly developed by Berthet-Rayne et al. [9], in which the trunk of the snake was separated into four lumens by four primary backbones for camera and instruments, just as Fig. 1.2D shows. Thakkar et al. [14] developed a snake robot system for MIS using the inner and outer concentric tubes driven by corresponding wires and feeders.

Table 1.2

ENT, Ear-nose-throat; NOTES, natural orifice transluminal endoscopic surgery; N/A, not applicable.

Figure 1.2 Mechanism classifications of tendon-driven surgical manipulators. (A) One primary backbone: tendon-holder variations. (B) One primary backbone: tendon variations. (C) No primary backbone: side view (left) and front view (right). (D) Four primary backbones (cross-section).

Meanwhile, a few of the snake robots adopted tendon-driven actuation in articulated modules, such as the shape memory alloy (SMA)-actuated neurosurgical robot designed by Ho et al. [20], and in vivo biopsy robot designed by Garg et al. [1].

1.2.2.2 Motor-actuated articulated snake-like mechanism

Articulated rigid-link snake robots with motors in their joints were relatively less reported than tendon-driven ones. Kwok et al. [21] developed a snake robot owning 10 DOFs with five joints that were actuated by micro-motors for MIS. Harada et al. [22] proposed a motor-actuated modular snake-like robot that can be self-reconfigured in the human stomach, plus the modules have been built for preliminary validations. Cepolina and Zoppi [23] developed a 6-DOF arm that can output accurate position and force for MIS. Omisore et al. [24] built a redundant four lead-module snake-like robot for abdominal radiosurgery. Son et al. [25] developed a semiautomatic snake robot that has 11 motor-actuated modules with rotating axis perpendicularly arrayed for transanal NOTES.

1.2.3 Novel design and platforms

Besides the conventional realization of snake-like robots such as tendon-driven and motor-actuated module connection, researchers have further explored new mechanical designs and actuation. As for the mechanism design, Conrad et al. [26] and Conrad and Zinn [27] developed an interleaved continuum-rigid manipulator that combines the advantages of precise actuation and safe contact. Lee et al. [28] used movable pulleys to drive 12 perpendicular joints of a snake-mechanism for NOTES. Dong et al. [29] proposed a twin actuation mechanism with a compliant joint that can maintain the cable tension in any configuration. Yoon et al. [30] developed a robotic system for maxillary sinus surgery using a compliant spring backbone to enhance interaction safety.

In actuation, pneumatic driving, as well as new materials such as artificial muscles and SMA, was used. Haraguchi et al. [31] developed a surgical manipulator driven by four pneumatic cylinders to realize force sensing without sensor fixed on the tip of the tool. Smoljkic et al. [32] proposed a hybrid rigid-continuum robotic system for MIS with the continuum part comprising an outer backbone of Ni-Ti tube and four McKibben muscles affording the bend motion. Lau et al. [33] built a robotic surgical system for endoscopic submucosal dissection with the continuum part comprising of a sheath and four SMA wires.

Soft robotics plays an essential role in the realization of novel snake-like surgical robots. Granular jamming and McKibben actuation were used by Jiang et al. [34] to build a soft snake robot with variable stiffness and flexible motion capability. Kim et al. [35] proposed a layer jamming tubular mechanism controlled by confining pressure for MIS. Abidi et al. [36] used fluid chambers to provide robot modules with high dexterity and safe interactions with human tissue. Wang et al. [37] built a soft surgical robot for cardiothoracic endoscopic surgery with a silicone-made manipulator driven by cables.

1.3 Modeling of snake-like surgical robots

This section briefly introduces the kinematic and inverse kinematic modeling methods based on the recent platforms that have been illustrated in the earlier sections. Moreover, statics is summarized with emphasis on the relationship between actuation force and deformation. Research on motion compensation, which is particularly essential for describing and solving the common problems that exist in snake-like robots caused by backlash, hysteresis, and friction force influences, will be concluded in 1.3.3.

1.3.1 Kinematics

The kinematic model of snake-like surgical robots has been intensively surveyed by Burgner-Kahrs et al. [2] and Li et al. [38]. As a complementary, here we will summarize the kinematic modeling methods of the platforms that emerged in the recent 3–5 years. As is known to all, the earliest modeling of the continuum robot configuration stemmed from the backbone curve proposed by Chirikjian [39]. Most of the snake-like surgical robots, especially the cable-driven ones, were modeled based on the piecewise constant curvature assumption. In the constant curvature model, once the length of the backbones of the robot is known, the configuration depends on the bending angle of the tip and the rotating angle of the bending plane with respect to the plane that is defined by the base disk. Simaan et al. [40], Kato et al. [41], Haraguchi et al. [31], Ding et al. [42,43], Li et al. [10,20,38,44–46], Smoljkic et al. [32], Lau et al. [33], Qi et al. [47], and Roy et al. [48] have validated their platforms using the kinematic modeling based on constant curvature assumption.

The conventional D-H method for rigid-link robots has been widely used too. Garg et al. [1] adopted the D-H parameter table on a wire-driven rigid-link snake robot to build its kinematic model. For a notched dexterous continuum manipulator in the work of Gao et al. [49,50], the interconnected rigid links and flexible links, transformation matrices were obtained respectively and construct the whole kinematic model–based on D-H method. Murphy et al. [51] proposed a wire-driven continuum robot composed of two nested tubes and used the single-chain D-H method to build the kinematic model. As a notched and nested assembly, the snake-like robot’s kinematic model in the study of Kutzer et al. [52] was built by 35 parameters from the pin joints and vertebrae one by one.

1.3.2 Statics and dynamics

A snake-like surgical robot’s static modeling solves the relationship of the force, moment, and deformation. For tendon-driven snake-like surgical robots, statics is usually combined with kinematics when Cosserat Rod Theory is applied in the modeling. Based on Cosserat Rod Theory, Gao et al. [17] built a shape prediction model for a helical spring backboned snake robot, in which the deformation of the robot is related to the tendon force, friction force, and external forces. Lumped-parameter model is an alternative basis for static analysis, for example, Kato et al. [16] built the tension propagation model with friction between the wires and the robot body. The principle of virtual work was used to compute the actuation force on building a load transmission model in the work of Roy et al. [48]. Dong et al. [29] analyzed the cable tension and stiffness of a compliant joint backboned snake robot based on the Jacobian. A dynamic model to compensate for the uncertainty and asymmetry has been proposed by Haraguchi et al. [31] by defining the driving forces related to the bending angle, friction force, and elastic forces.

1.3.3 Hysteresis and compensation

Brought by the tendon-driven mechanism, the friction force between the backbones and robot body introduces asymmetric hysteresis displayed as backlash. Moreover, the elongation of the actuation backbone is introduced by friction force and actuation force [33]. The backlash and elongation of actuation wires are the main factors that influence the accurate modeling and control of a snake-like surgical robot. Camarillo et al. [53], Kesner and Howe [54], and Gao et al. [55] recorded the hysteresis of the manipulator tip trajectory in the simulation and experiments. For solving the problem of hysteresis, Kato et al. [41] built an extended forward kinematic mapping method to improve the prediction of the posture with hysteresis included. Another way of the solution is to compensate for the backlash. Kesner and Howe [54] built a model to predict the width of the backlash zone and proposed the compensation method based on the offset values of the backlash zone. For the backlash compensation, Xu and Simaan [3] implemented a simplified redundancy resolution to update the instantaneous kinematics of the robot by adding a modifier to the actuation parameter; Simaan et al. [40] proposed an actuation compensation for both of the joint space and configuration space, with the compensation parameter obtained by linear recursive estimation. For the compensation of elongation, Bajo et al. [12] proposed the actuation compensation based on the elastic parameter of the backbones and energy information of the robot. Lau et al. [33] built an elongation model of an SMA-actuated cable-driven robot and handled the elongation and backlash by modifying trajectories in one model.

1.4 Human–machine interaction

Aiming at MIS or NOTES, surgical robots have to go deep into the human body. For surgeons, access to and manipulation of the target area is through the transmission of the mechanisms, and the operation vision is enabled by endoscopic camera or imaging through radioactive methods. Surgeons make the operation plan and manipulate the robots according to the feedback information collected by the robot system or external imaging systems during surgery. Therefore the commands and feedback interchange between surgeons and robots are essential to ensure a successful surgery, which highlights the role of human–machine interaction in surgical robotic research and developments. For snake-like surgical robots, the human–machine interaction becomes even more challenging because hyperredundant body shape brings difficulty in locating, configuration derivation, and motion planning. Moreover, most of the snake robots adopt the wire-driven mechanical design, which complicates the modeling and control.

1.4.1 Shape/force sensing

Tracking the position of the end-effector and configuration of the robot is the premise of safe access to human anatomy. However, the narrow access path into the target area cannot afford enough space for active measurement. It only allows passive measurement, such as electromagnetic markers. Besides, registration of the position and configuration of the robot to the preoperative 3D organ model is essential for surgeons during an operation. Shi et al. [56] summarized the measurement methods of continuum surgical robots’ position and configuration. Up to date, there are mainly two kinds of methods to track the snake robot inside the human body, one is curve-based shape reconstruction, and the other is extended Kalman filter (EKF)-based shape estimation combining the kinematics. For the first type, Song et al. [57–59] proposed a tip tracking and shape sensing method for tendon-driven snake robot without relying on a mechanical model, in which the robot shape was reconstructed by a three-order Bezier curve fitting of the data collected through electromagnetic markers on the robot’s critical segments and tip. For the filtering method, Tully et al. [60–62] and Srivatsan et al. [63] proposed EKF-based prediction of the pose, position, and configuration of a follow-the-leader style snake surgical robot, with the information of kinematic model and one 5-DOF electromagnetic tracking sensor. In the subsequent work, Tully and Choset [64] built a constrained Kalman filtering algorithm to localize the robot and registration it to the predefined organ surface model following the requirements of contact detection in the surgery.

Force sensing appears to be more challenging because force sensors need power, communication, and occupy specific spaces if they are fixed on the snake robot’s segments and tips. There are mainly two ways on the force sensing of snake-like continuum surgical robots, one is to measure the forces through force sensors fixed on the robot such as fiber Bragg grating (FBG) sensor, and the other is a deduction of for robot’s tip force through the static and kinematic model. For force sensing with sensors, Shi et al. [56] have summarized the usage of FBG sensor and fiber optical sensor on force sensing and shape estimation of continuum surgical robots. Separated from sensor-based force sensing, the deduction of the robot tip force is called intrinsic force sensing, which can be divided into categories of deflection-based force sensing and actuation-based force sensing. For deflection-based force sensing, Rucker and Webster [65] built a force estimation algorithm based on EKF using the robot’s statics and kinematics and the measurements of the uncertain poses. For actuation-based force sensing, Xu and Simaan [66] proposed an intrinsic force sensing method, in which the tip force was solved based on singular value decomposition) of Jacobian mapping from configuration space to twist space. Black et al. [67] proposed a force sensing method based on the generalized Cosserat-rod-based kinetostatic model of a parallel continuum robot. Yuan et al. [68] proposed a force sensing method using pose and cable tension of the robot based on the kinematic-static model. For a pneumatic driven flexible distal joint, Haraguchi et al. [31] proposed sensing of the three-axis external force on a snake-like forceps tip by estimation with a translational deduction based on the dynamic and inverse kinematic models.

1.4.2 Motion planning

Mechanical designs of snake-like surgical robots determine their workspace, which is one of the criteria of the robot’s dexterity. A snake surgical robot’s workspace can be derived from a forward kinematical model or backward kinematical model. Once the workspace of a snake surgical robot is known, it is essential to plan the motion of the robot to reach the operational area and manipulate the target. The anatomy of the operating environment is hard to model, which brings complexity to the robot’s motion planning. Even if the organs and tissues can be reconstructed in advance, motion planning of the robot should be careful by considering tissue deformation and collision avoidance. For a snake robot with 20 linked sections for the exploration of osteolytic lesions, without modeling of the lesion’s cavity, Liu et al. [69] proposed the motion planning, including collision detection based on sensor and sampling. Omisore et al. [24] proposed an inverse kinematics (IK) method for the planning of the path, with collision detection and avoidance at the assistance of virtual points. Chen et al. [70] considered less sweep area and target reachability as the motion planning criteria and proposed safety-enhanced planning based on a dynamic neural model.

1.4.3 Control

The snake-like surgical robots own hyperredundancy and unique mechanisms. As a result, complexities in modeling and motion planning arise, as have been summarized in the above sections. Moreover, the environment of human anatomy is narrow, curved, and deformable and thus hard to be modeled, especially when the robot is interacting with it. The robot itself and the environment in which it operates both enhanced the difficulty in the control problem.

1.4.3.1 Controlling variables

Position, force, and stiffness are the main issues in controlling the snake-like surgical robots. Mostly motion control of the snake robots is designed by optimization under constraints such as interaction with human anatomy, for example, Sen et al. [13] proposed to control for an 11-DOF snake-like palpation robot based on optimization under constraints of joint position and velocity limits; Kwok et al. [71] derived the motion modeling of an articulated snake robot under dynamic active constraints including proximity query status, haptic information, and visual information, to optimize the configurations and realize control of human–robot interaction; Li et al. [10] proposed optimal control for snake surgical robot by pursuing the highest stiffness and minimal movement in inverse kinematical solutions; Smoljkic et al. [32] realized control of a flexible robot for MIS based on expression graph-based task controller framework by quadratic programming of constraints of the pose of the tip and shaft. Hybrid motion and force control by Bajo and Simaan [72] for a multi-backbone continuum robot was built in a control framework that was composed of two separate controllers for the motion and force, respectively, considering the online estimation of compliance force and motion solution in the configuration space.

1.4.3.2 Controllers and their evolution

Prevailing controllers are implemented by proportional-integral-derivative (PID), proportional-integral (PI), or Jacobian of the system with feedforward structure, for example, Conrad et al. [26] proposed a control framework for an interleaved continuum rigid manipulator with two separate controllers, while the flexible segment controller has a feedforward inverse kinematic conversion. Haraguchi et al. [31] realized position control of a pneumatically driven snake surgical robot based on a PID cascade controller with dynamic compensating as feedforward.

Recently novel controllers emerged, among which modeless (or data-driven) control occupies a prominent position. Visual servo is the most representative one of modeless (or data-driven) control, in which inverse kinematics is obtained from external variables such as position obtained by cameras, depending less on the input parameters of robots, for example, Wu et al. [73] proposed a hybrid control for the teleoperation of a snake robot based on visual servo; Ouyang et al. [18] realized the visual servo control of the motion on their newly designed snake-like manipulator; Yip and Camarillo [74] proposed a modeless control featured by Jacobian estimation during the robot motion, and the control strategy was realized by optimization of tensions and changes of the Jacobian.

As an evolution, a learning method has been used in modeless (or data-driven) control. Xu et al. [75] applied regression methods on learning the nonlinear inverse kinematic model of snake-like surgical robots; Lee et al. [76] proposed a generic control framework, which learns the inverse model through online training without structural parameters; Mahler et al. [77] used Gaussian process regression to learn a nonlinear kinematics with velocity as a feature in the