Robotic surgery started with orthopaedics more than 35 years ago by a veterinary surgeon at the University of California who experienced a disabling illness that eventually prevented intricate hand movements. A project and later a company were founded to have a robot perform the manual bone incisions planned beforehand. In 1985, the first experimental canine surgery with a robot was performed with a fast-rotating burr held by a robotic arm that was programmed to mill a proximal femur for a total hip arthroplasty (THA) stem. The system was called Robodoc and continued development for human joint arthroplasty from that time to the present day. In the 1990s, navigation systems emerged that did not involve robotics, and in the early 2000s, PSIs started for total knee arthroplasty (TKA). All of the aforementioned navigation systems have continued to evolve in parallel, and rely on similar basic concepts.

Computer-aided surgery (CAS) robotic and navigation techniques emerged to overcome some of the limitations of the conventional technique by helping to provide better alignment accuracy by reducing outliers and supposedly complications of TKA and THA. CAS systems are classified into the following major categories: robotics
(active and semiactive, differentiation described later), navigation systems, PSI, and modern utilitarian smart tools, which include augmented reality systems.

All of these systems rely on some fundamental core-enabling technologies, functions, and steps. The systems vary by which of those steps are done, how, and by whom. The following description of such building blocks is hoped to provide knowledge about semantics and principle of operation of CAS systems, providing a solid framework for future self-learning.

The distinction between image-guided and non-image-guided navigation or robotic system relates to the virtual anatomic bone model on which, for example, a joint arthroplasty is planned and performed. In image-guided or image-based systems, a truly patient-specific three-dimensional (3D) model is created from CT or MRI data of that patient. The process involves segmentation first, and then 3D reconstruction of the data. CT, MRI, and even radiographic data can be accessed or transferred from an imaging machine in the DICOM (Digital Imaging and Communications in Medicine) format. Segmentation is a technique used to computationally extract from the full imaging data, subsets (segments) of the data points to represent one or more bone or any other tissue segments. Bone, for example, has high density and may thus be represented in the CT case by all the points with Hounsfield unit levels above a certain threshold. Other less-dense tissues, scale to lower Hounsfield unit levels, and metallic implants, if present in the patient, would be on the other extreme. Reconstruction is then performed on segmented data to build virtual 3D models of various bone segments. These models can be volumetric form, a solid model of many tiny bricks called voxels. More common and computationally efficient are surface models, which are hollow, featuring the surface only and made up of interconnected flat triangles, the vertices of which originated from the points of the scanned image data.

Using the aforementioned image-based bone modeling, the resulting virtual bone models have been used with general or customized computer-aided design tools to prepare a surgical plan for the patient. The outcome of the process can be one or more PSIs in the form of slotted cutting blocks (eg, for TKA), typically built by additive manufacturing (3D printing) so that they fit in one way to the patient’s bone profile during surgery to facilitate bone resections. PSI is now a mature technology offered by most orthopaedic implant manufacturers. Its main advantage is that tracking or robotics are not required in the operating room, and only a few custom (disposable) alignment instruments are necessary instead of the hundreds in a typical conventional mechanical instrument set. However, the process has some workflow limitations in having to involve a commercial manufacturer in the surgical planning stage and making of the PSI blocks.

In non-image-based systems, a bone model is displayed, but it only mimics the patient’s bone extracted from a library (atlas) of other people’s bones available within a database on the system and adjusted to fit. A suitable version from the atlas is parametrized and morphed (scaled by adjusting length, slenderness, location of centers, bone ends, and the scaling done uniformly or nonuniformly) to closely approximate to that of the patient. Measurements during surgery on the patient help that morphing process, including the location and distance between the hip center and knee joint line, the most distal point on each condyle, the distance between the epicondyles, and others. This process is sometimes termed kinematic registration, in which the bone model is approximately scaled, and tracked motions of the relative pose (location and orientation) of different bone segments to each other help estimate the hip, knee, and ankle joint centers, from which a mechanical axis is determined in the virtual computer space based on the patient’s physical one in the operating room.

A central function by a robotic arm, and/or a navigation system, is the ability to have the robot end effector or a navigation pointer, or a navigated instrument, be localized in space. This includes the x-, y-, and z-axis coordinates of the origin of the component measured, as well as its orientation by three angles around three axes along these dimensions. The pose that results for the patient bone and those of other elements in the surgery field (eg, pointers, navigated jigs, robotically held instruments, navigated implant trials) are then all obtained at a given time (instant). The measurements are then repeated and all elements are tracked dynamically in space from 30 to 100 times per second with current technology.

The poses of the robotic arm end effector can be measured by the robotic arm using the onboard position and angle sensors at the robotic arm’s articulated joints. Or, the robotic end effector can be tracked by using an external tracking system, which also tracks the bone and any ancillary instruments. In tracking, typically two cameras are used, spaced apart at a small angle to each other, for stereotactic vision resembling
the configuration of human eyes. In this way, 3D localization (including depth) is achieved with the processing of instantaneous pairs of two-dimensional camera images. Most cameras are filtered for infrared light only to reduce the effects of busy optical backgrounds and noise. Magnetic tracking can also be used, but has been much less popular because of the inherent magnetic interference (noise) from metallic surgical instruments in a surgical field.

In optical or infrared tracking, the cameras detect active or passive markers, typically in a unique constellation (called a reference frame, base, or array) composed of three or more markers rigidly attached to each tracked element (bone, cutting block, etc). Active reference frame markers are powered by batteries and electronics, so each marker on them is actually a small infrared LED (light-emitting diode) light, sometimes pulsed in a unique identifiable sequence for that device. Passive markers are much simpler and look like silver spheres to reflect incoming infrared light emitted from sources on each camera. Of course, both require a clear line of sight from the cameras to each element tracked in the surgical scene.

This is the process of measuring and computing to relate in 3D space the computer system virtual bone model to the physical anatomy of the patient. It is also required to register each of the aforementioned tracked reference frames to the geometry of the element to which it is attached before the element can be tracked. Registration can be performed in many ways: the simplest is point-to-point, by relating preselected fiducial points one by one to their corresponding physical anatomic landmarks, or by surface-to-surface matching, which is performed by digitizing a physical surface and computing where it would match with highest correlation (least error) to an equivalent patch of surface of the bone’s virtual model. There are various other registration techniques beyond the scope of this chapter, and they are an area for continued innovation and progress.

As it became possible to build 3D virtual models of the patient’s (actual or morphed) bone anatomy, register, and track them in the 3D physical surgical space and track relative 3D positions of alignment jigs (eg, cutting blocks, or bone resection instruments), various computer-aided orthopaedic surgical systems emerged. If a computer is programmed to move an actuator (robotic arm) carrying a resection device to a desired accurate position to resect bone, then that would be a surgical robot. If the robot itself was programmed to move the resection device to desired cutting paths, this was called an active or semiautonomous robot (eg, Robodoc). However, a robotic arm can be equipped with force/moment (haptic) sensors (typically a load-cell with six degrees of freedom near a user handle at the end effector) to detect where a surgeon manually moves the robot’s end effector or surgical resection instrument (eg, a milling burr or an oscillating saw blade). The computer, through the robotic arm actuation, would then either allow or resist the surgeon’s desired motion based on whether it follows the presurgical plan along preapproved cutting regions. This variety is sometimes classified as a passive or semiactive robotic surgery system (eg, Mako by Stryker). If the 3D tracking software guides the user to place a cutting block without a robot, this has been called a navigation system. In the following sections, additional categories of robotics and navigation systems used in joint arthroplasty, based mainly on their basic principle of operation, are presented.

The earliest of orthopaedic surgery active robotic arms was the aforementioned ROBODOC (by Integrated Surgical Systems/CA). In the late 1990s until the early 2000s, “Caspar”(OrthoMaquet/Germany) was another widely publicized system. Both articulated robotic arms were large, clean-room versions reconfigured from the manufacturing industry. They accurately moved a fast-rotating cutting burr to remove bone following previously programmed cutting paths much like a computer numerical control (CNC) machining process. Termed active (semiautonomous) robotic systems, they brought much excitement at the time and some published studies were optimistic if not bullish. However, multicenter studies on ROBODOC hip replacement in the United States (136 hip joints—almost half robot vs half conventional control systems) showed longer surgical time and higher blood loss with the robot, which were attributed to the learning curve. In a series of 900 cases in Germany, the Harris Hip Score rose from 43.7 to 91.5, and the surgical time declined quickly from 240 minutes for the first case to 90 minutes. The system was described as safe and effective in producing radiographically superior implant fit and positioning while eliminating femoral fractures.² The ROBODOC robot was discontinued and replaced with a modern, updated version named the TSolution^® One Robot sold by THINK Surgical^®, Inc., Fremont, CA (Figure 2).

The aforementioned shortfalls were inevitably going to be addressed by integrating interdisciplinary innovations. A noncemented TKA tibial component fixation has been developed for younger and more active patients, and also for easier revision should the need arise later. This contemporary solution (Figure 3) combines a radical and innovative TKA tibial component design with additive manufacturing and robotics. A light, agile robotic arm with a fast-rotating burr is automatically programmed during planning to accurately mill the peripheral fixation fin channel. Here a smaller, lighter robot is harnessed to perform a much smaller function for insertion of an implant targeting specific clinical advantages.

In this category of orthopaedic surgical robotics, the surgeon’s hand manually maneuvers the end effector that carries a resection instrument such as a rotating burr, or, lately, bone saws for TKA or reamers for THA. As is standard, they are programmed with a surgical plan, typically a CT image-based bone model rendered on the computer screen together with a model of the burr or resection instrument tip. Between the surgeon’s hand and the end effector is a load cell with six degrees of freedom that senses the forces and moments in the directions of motion the surgeon wants to move the robotic arm end. The concept of a haptically guided, surgeon-constraining robot was vigorously commercialized in the United States (Mako). Mako had also eclipsed the earlier UK Acrobot system, which had been the first (late 1990s) innovation³ in this category and with published clinical results.⁴ Mako also started with
a rotating burr for bone resections for a unicondylar knee arthroplasty. It has recently progressed to control a reamer in hip arthroplasties and a burr to knees and finally holding an oscillating power saw at the end of the robot effector for knee arthroplasties. Again, the Mako user interface allows the surgeon to resect the surface. The modern system’s peripheral instruments help guide the surgeon for ligamentous tensioning and gap balancing.

A 2020 systematic review⁵ that included 21 published studies on unicompartmental knee arthroplasties performed using the Mako system showed short-term benefits compared with the conventional technique regarding implant alignment accuracy, soft-tissue balance, patient function scores and satisfaction, complication rates, and learning curve in short-term outcome. However, the authors stated that these findings could not yet be extrapolated for midterm and long-term outcomes.⁵ A 2022 study⁶ retrospectively reviewed data from 2,392 conventional TKAs by six high-volume surgeons matched 1:1 with robotic TKAs by another six high-volume surgeons who used the Mako system. Outcome measures included surgery time, hospital length of stay, total direct cost, 90-day complications, utilization of postacute services, and 30-day readmissions. Overall, the median length of stay was the same and no significant difference in 90-day complication rates was found. However, the robotic surgeries took more than 10% longer to perform than the conventional surgeries, and the median total direct cost per case was significantly greater at $11,615 and $8,674, respectively (P < 0.0001).⁶ These authors also speculated about the cost justification and left it pending if it would be offset by lower revision rates and/or improved functional results.

Surgical robots position cutting guides with slots through which oscillating saw blades are guided. One of the earliest was the Praxim/Omni bone-mounted version (OMNIBot by Corin). It is fixed to the side of the distal femur from inside the incision by using two pins. The imageless configuration (Figure 4) is used for TKA in conjunction with BalanceBot, a robotic ligament tensioning tool. The BalanceBot has two expandable thickness paddles, one under each knee condyle, inserted into the knee after the tibial resection, actuated to certain force levels, and measuring medial and lateral gaps. It aids gap balancing by quantifying the soft-tissue envelope throughout the range of knee motion before femoral resections. Clinical data with this system totaling more than 30,000 TKAs (10,000 with the ligament balancer) were reviewed in a 2021 study.⁷ The results showed some utility in surgical planning and modestly improved accuracy compared with conventional and even navigation technology. Mechanical alignment improvements of approximately 0.5° were typically achieved, with TKA survivorship of 99.26% for
one cohort over 3 years and another 766 TKA series with 99.4% survivorship over 6 years. Gap balancing assessments and effects on pain were also studied and showed favorable results.⁸,⁹

The ROSA^® Knee System (Zimmer Biomet) was introduced, and it is much larger and non-bone-mounted (Figure 5). The robotic arm is fixed to a large-wheeled base cabinet. A separate wheeled console augments the system with external infrared cameras for tracking of the end effector/cutting block (Figure 5). This robot obviously has more sturdy positioning of the cutting block, but the price is the much larger footprint. With eight orthopaedic surgeons using the ROSA^® Knee System to assist TKA surgery with three different implant designs resulted in angle inaccuracy of less than 1° and precision (scatter measured by standard deviation) and mean resection plane offset and scatter of less than 1 mm, all measured for verification by a different computer-aided system on the same surgeries.¹⁰

A natural progression from the aforementioned robotic systems was to have the robotic arm constrain the motion of the power instrument (sagittal saw) directly to move freely in a single desired plane. The recently launched VELYS^TM Robotic-Assisted Solution (DePuy Synthes) has two-wheeled cabinets and achieves this functionality with a relatively smaller robotic arm (Figure 6). Here, the overall (course) positioning of the robotic arm is done manually through adjustable/lockable hinges. The final precise positioning that defines the final plane of saw blade motion is performed by smaller motorized actuators closer to the robotic arm end effector. This combination modestly reduces the size and lightens the weight of the overall system with only a modest compromise in rigidity. The VELYS^TM Robotic-Assisted Solution is relatively new, and reasonable-length clinical results are sparse.

These are small, smart passive devices that provide information during a procedure (Figure 7). Compared with conventional mechanical instrumentation, most do not require any additional pins nor involve line-of-sight issues associated with navigation. Some are disposable single-use, sterile handheld instruments and some only partly so. Their operation is based on miniaturized accelerometry such as that of smartphones. Accelerometers measure inclination (angles) relative to the earth’s gravimetric and magnetic fields. These systems combine an electronic compass and

sensing of the direction of gravity, in three orthogonal axes, on one surgical instrument. Combined with the possibility of multiple components interacting wirelessly, these smart systems can help align fixtures and cutting blocks for implants.

Some modest cost and potential waste are inherent in the disposable single-use versions. Other limitations are their accuracy and modest response rate. These are not serious as their use in joint arthroplasty is in a quasi-steady-state fashion, with bones and instruments moveable but assumed stationary for any given instantaneous reading. The other, more serious limitation is the accelerometer’s inherent inability to measure position or distance. Only angles are navigated, and not positions. However, the same electronic device can include miniaturized load-cell force sensors, which can aid in flexion gap balancing and ligament tuning in TKA.

The most attractive aspect of the accelerometer-based systems is their essential utility compared with the large-console robotic and navigation systems. Because of the similarity of their function simply being attached to mechanical cutting blocks, they naturally require a shorter learning curve, and have almost no extra operating room footprint and no line-of-sight issues, therefore increasing their appeal. Finally, the functionalities of some of these systems can at least theoretically be integrated with bigger navigation and robotic systems. The obvious next step is integration of two systems in a unified user interface for easier overall surgery.