Next Article in Journal
DDPG-Based Adaptive Sliding Mode Control with Extended State Observer for Multibody Robot Systems
Previous Article in Journal
Minimum Energy Utilization Strategy for Fleet of Autonomous Robots in Urban Waste Management
Previous Article in Special Issue
COBOT Applications—Recent Advances and Challenges
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:

Industrial Robots in Mechanical Machining: Perspectives and Limitations

Mantas Makulavičius
Sigitas Petkevičius
Justė Rožėnė
Andrius Dzedzickis
* and
Vytautas Bučinskas
Department of Mechatronics, Robotics and Digital Manufacturing, Vilnius Gediminas Technical University, LT-10105 Vilnius, Lithuania
Authors to whom correspondence should be addressed.
Robotics 2023, 12(6), 160;
Submission received: 11 October 2023 / Revised: 17 November 2023 / Accepted: 22 November 2023 / Published: 24 November 2023
(This article belongs to the Special Issue The State-of-the-Art of Robotics in Europe)


Recently, the need to produce from soft materials or components in extra-large sizes has appeared, requiring special solutions that are affordable using industrial robots. Industrial robots are suitable for such tasks due to their flexibility, accuracy, and consistency in machining operations. However, robot implementation faces some limitations, such as a huge variety of materials and tools, low adaptability to environmental changes, flexibility issues, a complicated tool path preparation process, and challenges in quality control. Industrial robotics applications include cutting, milling, drilling, and grinding procedures on various materials, including metal, plastics, and wood. Advanced robotics technologies involve the latest advances in robotics, including integrating sophisticated control systems, sensors, data fusion techniques, and machine learning algorithms. These innovations enable robots to adapt better and interact with their environment, ultimately increasing their accuracy. The main focus of this study is to cover the most common industrial robotic machining processes and to identify how specific advanced technologies can improve their performance. In most of the studied literature, the primary research objective across all operations is to enhance the stiffness of the robotic arm’s structure. Some publications propose approaches for planning the robot’s posture or tool orientation. In contrast, others focus on optimizing machining parameters through the utilization of advanced control and computation, including machine learning methods with the integration of collected sensor data.

1. Introduction

Industrial robotics plays a significant role in enabling high accuracy and repeatability in the machining process. However, industrial robotic machining can encounter various challenges, such as material [1] and tool variations [2], flexibility and adaptability [3], programming complexity [4], and quality control [5]. In some cases, because of environmental disturbances, robot positioning accuracy needs to be improved during machining processes [6]. The example of industrial robotic machining adapting different sensors and control algorithms is shown in Figure 1.
Industrial robotic machining processes and techniques involve using such systems to perform cutting, milling, drilling, and grinding operations on various materials, such as metal, plastic, and wood [7]. In the literature, robotic machining is classified into low material removal rate (MRR), e.g., grinding, polishing, and deburring, and high MRR operations, e.g., milling and drilling. The literature covers various topics related to the robotic machining process, including issues of robot dynamics, articulated robot configuration, tool trajectory optimization, quality, monitoring, and positioning error compensation. Low MRR research focuses on improving machining quality, while in high MRR operations, stiffness enhancement and vibration suppression are prioritized. However, no work has been reported to ensure stable and high-quality machining.
Advanced robotic techniques, such as force control or adaptive machining, optimize performance and ensure efficient material removal and machining [8]. There are challenges of limited machining accuracy with industrial robots and the need for improved calibration and compliance error compensation methods. In recent research, those challenges can be overcome using auxiliary units, e.g., additional actuation systems or external measurement systems, to improve the performance and quality of robotic machining overall systems.
Robotic machining is a feasible alternative to conventional computer numerical control (CNC) machines for processing materials with different levels of hardness and complexity [9]. However, the lack of rules guiding the use of robotic cells is a barrier that needs to be overcome. Some propose addressing the accuracy and quality of surface finish issues in robotic machining by studying the relative orientation of the cutting force concerning the robot’s stiffness.
The need is high for modeling and identification to optimize, plan, and control the machining process [10]. Recent research on various machining processes includes deburring, milling, polishing, and thin-wall machining. One major limitation of robots in machining tasks is insufficient rigidity at the tool center point (TCP), which affects machining accuracy. There are some suggestions for improving robot accuracy, such as implementing intuitive programmable systems, using sensors to adapt motions to the machining task, optimizing the path planning process, and offering cloud-based machine learning algorithms for process optimization.
Most of the inaccuracies arise from the dynamic properties of the specific industrial robot [11], tool path [12], machined material [13], or machined surface [14]. Hight vibration can occur during robotic machining, leading to a poor surface finish, reduced accuracy, and tool breakage.
This review aims to systematize existing knowledge and a multi-criteria analysis is conducted of existing methods and technological solutions in robotic machining facing dynamics problems, covering such issues as the stiffness-to-accuracy ratio, accuracy improvement methods, process optimization, adaptive feedback methods, robotic control, sensing, and machine learning algorithms.
The qualitative evaluation of papers dedicated to robotic machining cases, robotic tool path and posture planning, advanced robotic technologies such as control, sensing, and machine learning adaptations, and robotic material processing technologies is also performed.
The structure of this manuscript is as follows: Section 2 describes the methodology chosen for the inclusion or exclusion of articles for review; Section 3 provides an analysis of recent advances in robotic material processing covering milling, grinding, polishing, drilling, and some other cases; Section 4 presents a short review of industrial robot path and posture planning issues; and Section 5 covers some of the latest research on advanced robotic technologies, including control, sensing, and machine learning solutions. The manuscript ends with the discussion and conclusion in Section 6 and Section 7, respectively.

2. Method for the Selection Process

Different databases such as MDPI, IEEE Xplore, Science Direct, and Google Scholar were utilized. Semantic Scholar, an AI-powered research tool, was also explored because of several limitations (e.g., the article is only accessed in a specific database) after analyzing reference lists. Further, the selection approach for this manuscript was implemented using industrial-robot-related keywords, such as “industrial robot”, “robotic machining”, “robotic manufacturing”, “robotic milling”, “robotic grinding”, “robotic polishing”, “robotic drilling”, “robotic cutting”, “robotic sensing”, “robotic control”, “robotic path”, “robotic posture”, “robotic learning”.
Then, the titles and abstracts were screened using several criteria for articles that were defined for inclusion in this survey, such as:
  • is focused on industrial robots;
  • is focused on robotic posture and path planning;
  • is focused on machining processes where forces act;
  • is focused on sensors and their data fusion;
  • is focused on robotic control;
  • is focused on machine learning adaptations.
Correspondingly, defined exclusion criteria are:
  • articles older than five years are excluded, with some exceptions after reviewing reference lists;
  • the articles are not fully accessible;
  • articles not specifically focusing on industrial robots or data gathering were not selected;
  • articles providing information about industrial robot manipulation and packing;
  • articles providing insufficient data related to the manuscript.
Finally, repeated revision was performed to include additional articles based on the same criteria. The complete simplified selection procedure is shown in Figure 2.

3. Robotic Material Processing

Robotic material processing refers to the use of robots in various industrial areas for additive and subtracting manufacturing processes, i.e., milling, grinding, drilling, polishing, etc., instead of traditional machining centers. Different processing operations and parts implementation areas have unique machining accuracy and efficiency requirements. Figure 3 shows the graphical summary of analyzed use cases regarding the material, machining operation type, and focus point of the performed research.
From Figure 3, it can be concluded that different materials were used in different robotic machining research cases to improve industrial robotic posture, path, accuracy, or stability. In most cases, aluminum alloys are used as lightweight, corrosion-resistant, and desirable strength metals. In further sub-sections, the manuscript review continues based on the type of robotic machining operation.

3.1. Milling

Robotic milling offers cheap, effective, and accurate shaping operations, but in some cases, it has some limitations and requires intelligent solutions to overcome them (Table 1). The common problem in robotic milling operations is the complex relations between robots’ payload stiffness and accuracy. Robots with higher payloads are stiffer, but at the same time, they typically are less accurate. In addition, the stiffness of industrial robots is an extremely complex parameter depending on the majority of factors, including robot links configuration, the direction of gravity force, TCP position and orientation in the workspace, rigidity of each joint, etc.
Performed analysis of the reference outlines shows that that robots are more likely to chatter than regular machines due to lower stiffness [15], and active force control can be used to prevent this [16]. The enhancement of overall robot stiffness due to drives, body structure design, and materials is not actively ongoing.
Table 1. Summary of robotic milling.
Table 1. Summary of robotic milling.
ProblemMethods Industrial RobotAdditional MeansRef.
Low stiffness of the robot structureImpact tests at many tool-tip positions to obtain modal
parameters of frequency response functions (FRFs)
ABB IRB 6660-205Sound microphone, National Instruments (NI) data acquisition system[17]
Optimal control for active vibration suppressionLinear Quadratic Regulator (LQR) optimal control KUKA KR500-3 with KUKA KRC4 controllerKUKA Robotic Sensor
Interface and EtherCAT protocol, laser tracker
Machining accuracy based on stiffness properties Task-dependent performance index (PI)ABB IRB 6660-205NI data acquisition
the system, laser tracker,
Stiffness increase and
machining accuracy
Conversion from a 5-axis CNC tool path to a 6-axis industrial robot trajectoryMotoman MH80 IRAT960 laser tracker,
NX 8.5 software,
Leitz PMM-XI8106
Avoidance of over-cut and
Fixed cutter axis control
Motoman-UP6MATLAB-based design and simulation toolbox and ROTSY 4.2 software[21]
Prediction of surface
Mapping-based intersecting methodABB IRB 6660-205Kistler dynamometer, microphone, acceleration sensor.[14]
The insufficient stiff robot structure directly impacts the milling process. It defines acceptable cutting forces. Low stiffness and low cutting force of the cutting tool can excite vibrations affecting milling, accuracy, and surface quality [17]. This paper provides a method to predict the stability of the robot manipulator during milling, which is posture-dependent within the work volume. The approach involves conducting impact tests at many tool-tip positions to obtain frequency response at the corresponding robot postures and building a predictive model to construct milling stability lobe diagrams at different postures.
Another study [18] presents an optimal control approach to active vibration damping in robotic milling integrating a six-degree-of-freedom (DOF) industrial robot. The methodology involved using pose-dependent modal parameters to formulate and solve an LQR optimal control problem for active vibration damping. The methodology allows the reduction in vibrations in X and Z directions during milling and ensures lower deviations of the machined surface. According to the authors, future generations of industrial robots are supposed to minimize actuator delay so that they could make it better to actively control and enhance milling accuracy at higher frequencies of excitation.
An alternative approach evaluating the structural stiffness of an industrial robot to improve robot milling accuracy is presented in [19]. The authors introduced the task-dependent PI to assess the stiffness at a given robot posture. This index depends on the external acting force direction but is independent of its magnitude. In such cases, the machining process optimization goal is to maximize PI by adjusting tool orientation. A similar pose optimization method to increase stiffness and improve machining accuracy for milling robots is presented in [20]. The method involves the conversion of a five-axis CNC tool path to a six-axis robot trajectory, taking advantage of an additional DOF. A new frame-invariant PI is proposed to evaluate the stiffness of a robot at a particular posture, and a one-dimensional posture optimization problem is formulated considering joint limits, singularity avoidance, and trajectory smoothness. A simple discretization search algorithm that can be easily added to commercial CAD/CAM software solves the formulated optimization problem, providing a solution to ensure the highest stiffness in the required trajectory.
The paper [21] gives a solution for robotic milling surfaces of Non-Uniform Rational B-Splines (NURBS) with significantly large fluctuations, which is a usual task in the woodworking industry. The authors developed and validated the fixed cutter axis control method to avoid over-cutting problems that typically appear using the variable cutter axis control method. The required milling trajectory on a predefined NURBS surface is obtained using the NURBS mapping projection, evaluating surface curvature, and applying the equi-chord interpolation method to define a set of interpolation points.
A different approach to the increasing accuracy of robotic milling is demonstrated in research presented in [14]. The authors developed a model to predict the surface area topography of oriented Plexiglas in industrial robotic milling, taking account of vibrations from tools. They modeled the dynamic behavior of the intersection in the cutting zones and integrated the gained vibration displacements into the sweep surfaces of the cutting edges. These data were used to model the 3D graph representing the topography of the machined surface. Performed experiments showed that the predicted and physically measured surface topography, and the model predicting the surface topography inducing the regenerative chatter, agreed well.
From the review of the literature about robotic milling, the main objective is to improve accuracy from the perspective of robotic stiffness. There were several approaches to this, such as through vibration measurements, e.g., implementing impact tests to obtain modal parameters of FRFs and adjusting the performance to the data, or laser tracking, e.g., implementing the LQR optimal control method to suspend vibrations by monitoring the amplitude.

3.2. Grinding

Robotic grinding covers performed precision material removal processes, typically using abrasive tools or wheels, to achieve desired surface finishes or remove excess material from workpieces. Some research cases on robotic polishing are summarized in Table 2.
Robotic grinding is assumed to be an alternative approach in complex component machining because of its flexibility, intelligence, and cost-effectiveness [22]. The paper presents a systematic review of robotic grinding, focusing on accuracy, compliance, and cooperative control. Relevant research work and various strategies to overcome challenges are discussed here, including online measurement, allowance control, force control, and surface integrity. However, despite the technical advantages of robotic machining, there are shortcomings remaining in terms of accuracy and surface quality compared to CNC machine tools due to the high coupling of robotic performance and configuration. To further improve the robotic grinding performance capability, breakthroughs in robot calibration, monitoring, trajectory planning, force control, and surface consistency are urgently needed. According to the review authors, future studies should address parallel measurement accuracy and stability issues, splicing point cloud fragments without mark points, and using inspection software for surfaces with a common data-storage format and a unique library of validation processes.
The surface curvature characteristics of complex-shaped stone products (CSSPs) were examined, and the correlation between surface characteristics and machining trajectory was studied [23]. The optimization of grinding force variations in the finishing process was carried out, and the optimal machining trajectory was set to improve the surface profile tolerance. Results from simulations were then experimentally verified. It was found that changing the trajectory can reduce the grinding force variations by 52.8% and the surface profile error by 36.9%. It was concluded that the surface finishing and tool axis changes with the surface direction result in lower grinding force variations in stone. The optimization effect was the most significant when machining a concave cylindrical surface, with a 63.7% reduction in fluctuation.
Despite its potential, the quality of robotic grinding for free-form surface machining is limited by individual structural errors or joint stiffness [24]. A systematic method of error compensation, workpiece orientation optimization, and optimization of the tool pose is proposed in this article to reduce the machining error in robotic grinding, which is a viable technique for free-form surface machining. The machining error mathematical models are built using speed and force adjacent transformation, and an error compensation approach is presented by fine-tuning the workpiece position. The experiment of blade grinding is performed, and the proposed method successfully reduces the average machining error of the blade. This approach may be useful to stimulate the use of robotic belt grinding for parts with high machining quality requirements, taking into account two factors: kinematic tolerances and joint stiffness.
An adaptive trajectory planning algorithm is proposed, using an MRP model, for the robotic belt grinding of complex blades in this paper [1]. The algorithm takes into account the elastic deformation at the contact wheel–workpiece interface and the curvature change characteristics of the free-form surface to enhance profile accuracy. According to results of simulations and experiments, the algorithm effectively solves the problem of over-cutting while reducing machining time and improving surface roughness. The average profile error is only 0.0194 mm and is within the tolerance of 0.08 mm. The controlled contact force and a variable process parameter approach suit the grinding of free-form surfaces.
In the paper from above [25], the TVIS as a virtual surface used to generate constant force in robotic grinding of weak-stiffness workpieces was defined, including contact trial (CT) and the reconstruction of the surface. The CT process measures the deformation and stiffness of the workpiece, and a TVIS mesh is constructed for grinding path planning. Experiments show that the proposed method achieves constant grinding force and is robust to different types of workpieces and techniques. The paper’s main contributions are the automatic method to realize constant robotic grinding force, the TVIS concept for constant force grinding, and the CT process to measure the workpiece deformation and stiffness.
A robotic grinding workstation designed for precision and automation in grinding rough metal cast objects is presented in [26]. The system employs machine vision and an industrial manipulator to address challenges in positioning and automatic grinding. The approach includes a two-step localization strategy, combining deep neural networks and template matching for precise object positioning in challenging industrial environments. Edge extraction and contour fitting techniques are used to locate and address surface imperfections, particularly burrs, on the objects. The system uses a grid method to detect and plan the grinding trajectory for the industrial manipulator, enhancing automation throughout the loading, grinding, and unloading processes. The results demonstrate the system’s stability, efficiency, and accuracy, while also overcoming material-related challenges. According to the authors, this robotic grinding station significantly improves precision, quality, and production efficiency, reducing manual labor and increasing staff safety.

3.3. Polishing

Robotic polishing involves using robotic systems and tools to achieve high-precision and efficient surface finishing, improving productivity and consistency in various industries as the final machining process. While the robotic arm allows flexible manipulation of the polishing tool, the stability of the manipulator arm is also affected by the acting forces of the tool. This impacts the machined surface quality, and force control is demanded. Several robotic polishing research cases are summarized in Table 3.
In the manufacturing industry, polishing is considered the final machining phase. In the research [27], process parameters (i.e., polishing pressure, belt speed, max cutting depth) were optimized for robotic polishing to reduce surface roughness, thus improving the quality of the mold’s steel surface. The optimum range for each parameter was obtained from a single-factor experiment. Another central composite design (COD) experiment was conducted to establish a prediction model of surface roughness. Here, the COD experiment can evaluate the linear and interactive effects, such as the high-order surface effect. The polishing experiments were conducted for validation, and the experimental results showed an effective decrease in surface roughness after polishing with the optimal parameters such as polishing pressure, tool rotation speed, and feed rate. The prediction model helps improve the quality of the surface in robotic polishing. Additionally, a platform with force control was built here to realize the robotic polishing of the workpiece.
Applying thermal barrier coatings in aerospace manufacturing for aero-engine blades allows them to operate at higher temperatures with greater efficiency. However, it increases the surface roughness, which affects the thermal lifecycle and insulation performance of the coating [28]. To reduce the roughness, an integrated robotic polishing trajectory planning method is proposed in this paper. The retrieved point cloud is preprocessed and, after that, the slicing and intersection methods are used to create a contact point set. The tool-tip positions and postures are calculated to optimize the robot’s movement, resulting in a smoother polishing trajectory. Simulation and experiments show that the coating roughness on the blade surface after polishing is reduced. Polishing is achieved with a small amount of removal, which becomes mirror-like smooth.
It is possible to use complex force-controlled end-effectors with few degrees of freedom to achieve consistent surface quality in robotic polishing. An end-effector with two rotational and one translational joints is presented in [29]. The simultaneous control of the angle of inclination of the polishing disc and the normal contact force between the disc and the surface of the workpiece is considered to be very important. The contact point variation, caused by the changes in the angle of inclination of the end-effector’s moving platform, the disc’s geometry, and the disc’s compliance, is identified as a significant issue that reduces the accuracy of force control. A combination of force control and contact point prediction results in a new hybrid orientation/force control architecture with a contact point compensation model used to control the orientation of the tool and the contact force. Such a method significantly improved the force-tracking performance under three different angle of inclination references, reducing the mean force error by 78.9%, 81.1%, and 72.3%, respectively.
The paper [30] proposes another constant force-tracking control scheme adapting an impedance controller in a combination of online stiffness and reverse damping force (OSRDF). The goal is to track the desired machining force. The proposed approach tracks and references the trajectory based on its reference position and velocity. This OSRDF controller is implemented by adjusting the stiffness parameter and merging the inverse damping force with the force tracking error. This is used to compensate for the properties of an unknown environment and reduce the force error to close to zero. Simulation studies and experimental tests on a robotic manipulator showed the efficiency of the proposed method.
Complex multi-degree-of-freedom tools certainly provide additional flexibility for controlling the machining process, but, on the other hand, they require more effort for accurate control. An alternative solution is passive devices maintaining the constant force between the tool and the machined object [31]. Using the constant force tool eliminates the requirement for force monitoring and real-time control. The passive constant force end-effector was designed to provide a 40 N constant contact force, while the polishing experiment was conducted on a reflective mirror 200 mm in diameter. This showed that the passive end-effector provides stable pressure with variations within 3.43 N. The polished surface root mean square of a large-aperture reflective mirror is lower than λ/10 (where λ = 632.8 nm). Compared to traditional force control methods, the passive method provides a low-cost and trustworthy choice for robotic polishing using a constant force-compliant tool. Nevertheless, issues of the mechanisms’ stiffness remain essential points.
An original passive end-effector design based on a constant force mechanism for robotic polishing was proposed in [32]. The end-effector regulates the contact force passively, and the motion range acts as a buffer to counteract excessive displacement induced by inertia. With no force overshoot, the surface quality of the workpiece is consistent. The experiment results show that the constant force mechanism improves the force accuracy and consistency of the polished surface.
The article [33] describes a technology to improve finishing efficiency, surface quality, and surface consistency for robot automatic polishing on curved surfaces. A polishing platform was created, and a study established a pattern between removal rate and polishing pressure. A polishing pressure control model was also established, and factors influencing polishing were discussed. The article proposes a generation algorithm for the position and posture of the robot polishing tool, which uses force–position–posture decouple control. This is used to achieve robotic polishing that is similar to that experienced with manual constant pressure polishing. It was tested for automatic robotic polishing on bent surfaces, and the validation results showed the efficiency and feasibility of the model, as well as its ability to achieve precise surfaces. The article also introduces a gravity compensation algorithm for the polishing tool to eliminate interference caused by gravity during machining.
To summarize, there are two main objects of research in robotic polishing: surface roughness reduction and improvement in surface quality, and constant force tracking between the robot tool and the machined workpiece. One can be achieved by optimizing process parameters through force measurements or scanning the 3D profile of the surface. The other can be achieved through impedance control or a passive constant force end-effector.

3.4. Drilling

Robotic drilling refers to using a robotic system to perform drilling operations, offering accuracy, efficiency, flexibility, and increased safety in various manufacturing, construction, or aerospace industries. From the perspective of robotics, it faces similar issues to milling but raises additional requirements for tool orientation—the tool rotation axis must be kept at a constant angle during the procedure. Therefore, the main challenges for such operation are non-uniform robot stiffness and the impact of gravitational force limiting the implementation of robotic systems in some industrial areas requiring high accuracy.
The use of six-axis articulated industrial robots for drilling and trimming in aerospace manufacturing is increasing due to their flexibility, low cost, and large operating volume [34]. However, too low structural robot stiffness and positioning accuracy limit their machining to non-critical components and parts with low accuracy requirements for surface finishing. A study was conducted to improve the capability of robotic machines to compensate for robot path errors in real time using a single 3-DOF laser tracker. The experiment significantly improved path accuracy, hole position accuracy and quality, and machined aluminum part accuracy. However, the study also showed that when compensating for the robot in real time, the largest source of residual error remains the backlash.
On the basis of vibration, a classification approach was published to detect low-quality holes caused by robotic inclined drilling in the aircraft assembly industry [35]. The approach involves establishing a vibration model to simulate its signals during vertical and inclined drilling and training a Resnet classifier using actual and simulated signals. In that way, the drilling state is identified by inputting the signals of the intercepted stable drilling section into the Resnet. The proposed method greatly reduces manual operations and solves the problem of obtaining many set data to train deep learning methods. From the experimental results, the Resnet classifier can accurately classify vertical and inclined drilling, even if few actual signals are used.
Another analysis of robotic drilling quality, which is influenced by its stiffness, is given in [36]. From here, drilling thrust force can cause the deformation in three directions of the manipulator, resulting in low diameter accuracy and poor hole quality. The paper proposes using a preload pressing force with a defined stiffness promotion coefficient to strengthen the stiffness of the machining plane. A model is established and the effects of pressing force are quantitatively evaluated. A matching criterion is proposed to ensure stability during drilling and, based on this, the proper pressing force value can be calculated. The experiments showed that the machining plane stiffness could be effectively improved under preload pressing force, leading to enhanced drilling stability and precision of the hole diameter.
Another study [37] focused on accurate drilling applications, and found that a key source of errors affecting drilled holes’ quality and circularity is static friction in robot joints, especially when joints reverse direction. The study proposed an optimization framework to improve robot manipulation and achieve better machined properties of the hole. The proposed model includes a general robotic drilling manipulation with an additional DOF. Particle Swarm Optimization (PSO) is used to select the best pose throughout the motion, and experimental tests show the tool deviation envelope is reduced by 40%.
The possibilities of an industrial robot drilling with a high-speed machine tool spindle for machining aluminum 6061-T6 were investigated in a study [38]. The hole exit burr heights and surface roughness were assessed by implementing the Taguchi design methodology, focusing on the feed rate, spindle speed, and pecking cycle. A condition monitoring system was utilized to identify the vibrations during the drilling operation and to determine which robot poses have increased stiffness. According to the authors, there is potential for further improvements, in particular with control of the drill exit force and the pecking cycles.
Recent advantages in the virtual and augmented reality fields also have a corresponding response in robotics. A digital twin framework for robotic drilling is introduced in [39]. This is proposed based on the idea, where three entities have been defined: the device communication entity, the digital twin entity, and the user entity. The authors present a developed generic reference model to highlight elements of the digital twin architecture relevant to robotic drilling, and real-time visualization of drilling process parameters is demonstrated. According to the authors, this framework for robotized drilling may be used for virtual real-time 3D visual representations. However, developing a comprehensive digital twin covering all aspects of the robotic drilling process is costly and time consuming. However, on the other hand, such a digital twin complemented by artificial intelligence methods could be used to predict system behavior or optimize the efficiency of the process.
A summary of recent implementations and research on robotic drilling is presented in Table 4.
In conclusion, stiffness is a research target that is similar to previous targets in the field of robotic drilling, and another research target is drilled hole quality. Some propose using preload pressing force, whereas others simulate vibration signals and train a Resnet classifier using actual and simulated signals identifying the drilling states, thus minimizing the resources needed for computation. For example, a study has been carried out by comparing CNC drilling and robotic drilling, and showed that using the Taguchi design methodology for the latter can improve machining quality.

3.5. Other Cases

Robotic material processing is not limited only to traditional machining operations. There are a lot of other less used cases, such as chamfering, deburring, or boring, which require robotic tool path accuracy to achieve the best quality and efficiency. According to [40], robotic deburring is a suitable and cheaper alternative because of its flexibility and the use of laser, vision, and force control sensors. In addition, Ref. [41] puts forward a robotic tool study as a key factor for a better chamfering quality. The summary of some of the most interesting cases is presented in Table 5.
The paper [42] describes a method for robotic gear chamfering that avoids the use of expensive and time-consuming measuring devices for more accurate positioning, registration, and fixing of the workpiece. Because of the uncertainty of the workpiece orientation and location, authors employ chamfering trajectory generation and force together with motion control strategies to identify the center and root positions of the gear. The method uses a compliant device, which can help to keep contact between the tool and workpiece edges and also helps to keep the force along the axis. The results showed that the proposed system has an effect in terms of chamfer machined quality and chamfer machined width, setup time, and registration time reduction. The paper also presents the relationship between compliant force and chamfer width at a fixed spindle speed of 5250 RPM.
Another study focused on robotic deburring and intended to compare the two methods’ performances to identify the suitable predicted forces during the machining of parts having areas of less than 1 mm [43]. One method used for the research is the linear regression method and the other is simplex; these methods were used based on measured machining parameters, such as spindle speed, feed rate, feed per tooth (FPT), and cut depths (tool immersions) to produce force estimates. Tests were performed by implementing two sets, where in the first one the FPT was constant and in the second the FPT was linearly increased. Experiments were performed in practice with a six-DOF robot and forces were measured with a force–torque sensor, which was mounted between the last robot joint and the spindle. Both methods demonstrated successful prediction results—the linear method was four times more accurate in full tool immersion, whereas the exponential model showed better results, with a material removal that was eight times lower.
Ref. [44] introduces a new method for identifying and predicting chatter in an industrial robotic boring system. Chatter is a common issue that affects the quality and efficiency of the boring process. The method involves three steps. First, the two decompositions of the measured force signal into intrinsic mode functions (IMFs) and a residue using the empirical mode are performed. Then, the Hilbert transform is applied to each IMF to obtain instantaneous frequencies and magnitudes, creating a Hilbert–Huang spectrum of the original signal. Lastly, chatter features are extracted by analyzing the Hilbert spectrum of each IMF and chatter symptoms are detected using statistical methods. Experimental results demonstrate that this approach can predict chatter up to 0.6 s before it occurs, aiding in subsequent chatter suppression and improving the workpiece’s surface quality. Additionally, the study found that the vibration frequency in a robotic boring system varies from low to high when chatter occurs, unlike standard machine tool systems where the frequency changes from high to low.
There are, of course, many other applications, but they were not included in this review because the publications were either older or did not meet other selection criteria, such as publishing date and quality.

4. Robot Path and Posture Planning

4.1. Robot Path Planning

Tool path planning in industrial robotics typically means generating optimal paths to follow with the end-effector while performing a specific task. Obstacle avoidance, path smoothness, and travel time minimization are considered to optimize the robot’s movements. The summary of typical problems and methods related to tool path planning is presented in Table 6.
As seen in Table 6, insufficient manipulator stiffness is a significant issue related to tool path planning. Ref. [45] described the challenges caused by insufficient manipulator stiffness of industrial robots when they are used in high-accuracy machining of large-scale components (Figure 4). Errors in the position of the motor shaft and linkage caused by joint flexibility cause the robot to deviate from the desired path. To overcome this issue, a real-time path correction based on JPE estimation and compensation was presented. That approach uses a link state estimator called FDBDKF to estimate the JPE and is based on a local weighted projection regression scheme to predict and compensate for the future JPE. Simulation and experimental results demonstrate the feasibility and effectiveness of the proposed approach. It shows a significant improvement, by more than 80%, in the path accuracy of a machining process.
M. Cordes and W. Hintze [46] conducted a study where they presented a model predicting the path deviation of industrial robots in joint space according to joint stiffness and reversal error. This model was validated by milling circular contours and evaluating hysteresis at the TCP trajectory. It was found that path deflection of the TCP was mainly because of the limited torsional joints’ stiffness, and errors increased linearly with the lever arm. At low cutting forces, particularly at joint 1, the hysteresis due to the lack of gravitational preload significantly increased the overall error. The model parameters were identified and confirmed by measuring hysteresis at the TCP, and roundness was improved by 31%.
An alternative solution for insufficient manipulator stiffness in the case of lightweight parts is the change between Tool-to-Part (TtP) and Part-to-Tool (PtT) concepts. M. Stepputat et al. [47] presented an approach using a low-load robot equipped with a gripper and programmed with an APT code to generate a PtT path to handle all production steps for the production of customized small wooden parts. In the PtT scenario, the robot handles the material rather than the tool, allowing the part to be moved in all six directions. Thus, the system enables a more accessible automatic material supply, manufacturing stations, and dispensers to take processed parts. The user interface can guide untrained personnel, and the approach allows for freehand entries to be milled automatically. The article highlights that the proposed approach reduces the human resources required for manufacturing.
Low stiffness of the workpiece also could significantly impact the overall machining process. In [25], a novel robotic grinding method was proposed to address the issue of serious deformations and vibration during the grinding of weak-stiffness machined parts. A TVIS was defined as a virtual surface to generate constant force during grinding, and a CT process was employed to actively sample the deformation and stiffness of the contact point using a force sensor. The authors included a TVIS concept for constant force grinding and an automated robotic approach that does not rely on high-precision vision instruments or modeling of the tool, workpiece, and fixture. The suggested approach was robust to different types of workpieces and processing methods.
Related research is provided in [1], where an iso-scallop height algorithm is proposed for robotic belt grinding. The trajectory planning method is considered to be a significant factor resulting in poor accuracy of blade edges. The proposed algorithm, according to the MRP model, with regard to the elastic deformation at the contact wheel–workpiece interface adapts grinding points which are based on the curvature change characteristics of the free-form surface. The proposed algorithm effectively solves the problem of over-cutting, ensuring an average profile error of 0.0194 mm of the entire blade and reducing the machining time by 68%. A method for minimizing energy in manipulator trajectory planning for high-speed machining of sculptured surfaces is proposed in [48]. The robot machining system’s energy characteristic model is set up to acquire the energy-optimal feed rate, and a planning model with complex constraints is created. The suggested strategy minimally modifies the initial objective-optimal B-spline feed rate curve (BFC) from the original trajectory planning. The proposed method significantly improves the efficiency of robot trajectory planning while exhibiting excellent performance.
Ref. [2] discusses a time-optimal motion planning method for robotically machined sculptured surfaces, taking into account tool-tip kinematic limits and curved tool path complexity. The method considers joint space and tool-tip kinematic constraints. It uses a numerical integration method based on the Pontryagin maximum principle [54] to solve the problem. The algorithm is implemented on the ROS, and experimental results validate its advantages compared to conventional motion planning algorithms. The proposed method simplifies and decouples the constraints, reducing calculating complexity and making it more favorable for the surface machining process.
Ref. [53] discusses the development of a scanned path generation for robotic non-destructive testing (NDT) of complex-shaped workpieces. Using the swept frequency eddy currents method, the researchers used a programmed off-line scanning technique to inspect the fixed leading-edge panel of an aluminum aircraft wing. They reconstructed a CAD 3D model of the surface through reverse engineering. The six-axis robotic arm KUKA KR5 arc was utilized to deploy the eddy current probe. The paper reports that the positional uncertainty of the NDT scan does not exceed 0.5 mm, which is moderate considering the uncertainties related to the off-line robot programming.
To summarize, there are a number of approaches to improving the robotic path during machining. One approach is to stiffen the industrial robot, minimizing the vibration impact on the process. This can be achieved by changing the robot joint position and analyzing the system vibration with force sensors. Another approach is to identify factors affecting the machining path, and to correct it or predict and generate the optimal path. One of the approaches to implement it is to adapt the path scanning using laser or eddy currents, which can simplify the problem solution.

4.2. Robot Posture Planning

Posture planning in industrial robotics involves determining the optimal configuration and orientation of joints and the end-effector to achieve accurate and efficient task execution. Several actual research cases focused on posture planning are summarized in Table 7.
Ref. [11] established guidelines for selecting the appropriate static and dynamic robot stiffness models based on the robot and process characteristics. These are used to optimize the robot pose minimizing the end-effector displacement. It was found that a static model may be sufficient for pose selection for tasks where the process forces caused by vibration frequencies do not approach the robot’s resonant frequencies. The dynamic-based pose selection performs significantly better in reducing end-effector vibration when the process forces excite vibration frequencies close to the robot’s natural frequencies. The study revealed the advantages and drawbacks of both dynamic and static models. According to the authors, the best possible results before optimization are provided when all other parameters affecting the process forces and the robot stiffness have been selected.
Research in [17] presents an approach to predict industrial manipulator stability at any posture based on the TCP FRF, which is dependent on a posture within the work volume (Figure 5). Many TCP positions were oriented to conduct the impacting tests, and the FRFs of all tested robot postures were identified. The inverse distance-weighted model was used to predict the TCP FRF at any position when obtaining the model parameters. The feasibility of the approach was proven by performing robot machining experiments verifying the stability lobe diagrams. The paper concludes that utilizing a suitable set of robotic machining parameters considering the predicted stability allows the prevention of the chatter of the tool.
A study in [19] was conducted to analyze the stiffness properties of a manipulator to improve the accuracy of robotic machining and to establish optimization methods for its posture and tool orientation. A task-dependent performance index (NSPI) of robot stiffness was proposed to evaluate it in the surface normal direction of a machined part for a particular posture. A distribution guideline was also proposed for the NSPI concerning any direction in the Cartesian coordinate system for a particular posture, which revealed the anisotropic stiffness property of the robot. A model was established to optimize the posture of a robot in a machining application by maximizing the NSPI. During experiments, the feasibility of using the proposed model to improve machining process accuracy was validated.
A similar approach in [55] presents a methodology for posture optimization of an industrial robot based on PIs. It proposes to consider regions of the robot workspace while performing machining. Regions where kinematic, static, and dynamic performances are at their highest allow machining errors to be reduced and best utilize the advantages. An optimal initial placement of the part, considering the robot’s properties, can be achieved by optimizing the robot’s performance. Performance maps drawn according to kinematic, stiffness, and deformation evaluation indexes refine the robot workspace, and the globally optimized robot posture can be derived. These evaluation indexes’ feasibility and posture optimization were validated by conducting experiments on a robot platform.
As is seen from the analyzed publications, robot posture planning mainly deals with the irregular stiffness of articulated robots. The development of models and maps allows the acquisition of optimal solutions for placing the workpiece or selecting the best robot configuration. Nevertheless, it should be noted that practical implementations of the mentioned methodologies strictly depend on the occupancy of the robot’s workspace and positioning task; a limited workspace and complex positioning tasks with exactly prescribed tool orientation result in lower optimization possibilities.

5. Advanced Robotic Technologies

Advanced robotic technologies cover cutting-edge developments in robotics, integrating advanced control systems, sensors and their data fusion, and machine learning (ML) algorithms to enable robots to adapt to and interact with their environments with enhanced precision. These technologies include such systems as computer vision for the object as a target or as an obstacle recognition, trajectory correction, or machining accuracy improvement.

5.1. Advanced Control

Sometimes industrial robotics requires advanced and unusual control systems to enhance robot operations within industrial settings. In modern robotic control systems, human–machine collaboration is increasingly important, and accurate payload estimation is necessary for such collaboration [56]. An identification method for a six-DOF robot was created to optimize, as the author called it, the excitation trajectory and to capture the underlying dynamics using fewer parameters. Here, the excitation trajectory is the optimized trajectory using fewer parameters. Problems including effective dynamic parameter identification and precision motion control with accurate payload estimation are addressed here. The simplified identification method with 35 parameters of the posterior essential parameter set achieves equivalent accuracy compared to the conventional method with 52 base parameters. Additionally, an integrated direct/indirect adaptive robust control (DIARC) algorithm was proposed. It consists of a generalized momentum-based indirect adaptation law for the payload estimation, immediate compensation of the adaptation transients, and robust feedback for the uncertain nonlinearities’ attenuation. Experiments showed payload estimation to be the actual value, and the proposed controller can achieve better tracking performance.
Ref. [57] presented an improvement in the machining accuracy of manipulators using a CNC controller as the control system. The influence of motion acceleration and running speed on path accuracy was analyzed by evaluating corner paths. The experimental analysis reveals that the controller provides a steady robot performance when running a corner path, as the constant running speed could keep a constant contact force and material removal rate. The conventional robot controller KR C4 is position oriented, while the CNC controller is path oriented and continuous. The researchers highlight the necessity to further develop robot controllers regarding the straight path besides adopting CNC, as robot manipulators are applied in machining processes.
A robotic system supplemented by augmented reality (AR) for trajectory interaction is presented in [4]. The system uses a mixed reality display mounted on the head and a robot arm to facilitate the robot programming task through four interactive functions: trajectory specification, visualization of robot motion and its parameters, and online reprogramming during simulation and execution. This robotic interface is validated by comparing it with a kinesthetic teaching interface. The paper also reports an industrial case study where the AR robotic interface reduces wrinkles during the pleating step of the carbon-fiber-reinforcement-polymer vacuum bagging process. The results showed that the AR robotic interface requires less teaching time and has a lower workload than kinesthetic teaching but presents an increased mental effort. The paper concludes by suggesting potential improvements for the system, such as introducing force visualization and control during robot programming, and more agile path specification methods.
A summary of research focused on smart control issues is presented in Table 8. It can be noted that additional controllers allow fostering the functionality and adaptability of industrial robotic systems, but on the other hand, such solutions result in other limitations. Adding new higher-level controllers increases system latencies and can only partially solve the problem of accurate motion control since using the regular robot controller, which is responsible for low-level drive control, is unavoidable in most industrial cases. Therefore, motion accuracy is typically increased only by minimizing distances between the specified points in the trajectory. Several force control methods have already been mentioned in previous chapters; therefore, they are not included here.

5.2. Robotic Sensing

Industrial robots, together with integrated sensing technologies, have significant advancements. These enable robots to possess enhanced perception and adaptability for the accuracy required in tasks by implementing vision systems, force/torque sensors, or proximity sensors. A few applications and research cases of robotic systems complemented by sensing technologies are presented in Table 9.
The increased usage of industrial robots in large-scale manufacturing has been driven by the high demand for efficiency and cost reduction [58]. However, the poor absolute accuracy of these robots during machining processes has been a significant disadvantage due to their serially connected links, the gearings’ resilience, and sensitivity to temperature changes. The research paper [58] proposes an approach to increase the overall positioning accuracy of an industrial robot for milling operations by using a laser tracker system to measure the TCP position and orientation. Position deviation is compensated for by implementing a supplementary controller in the manufacturer’s CNC, which allows the correction of the trajectory; thus, the machining path matches its specifications. Integrating real-time laser tracking in an adaptive milling work cell offers a highly efficient alternative for machining large-scale components, e.g., in the aerospace industry, where high tolerances of carbon-fiber-reinforced plastic components are required.
A technologically similar, model-based approach to calibrating woodworking machinery by measuring setup errors at the tool tip using a motion capturing system was proposed in [59]. The study focused on using this system for calibration by relating measured position error data with machine parameter values implementing the nonlinear least squares method. The effectiveness of the proposed system was demonstrated experimentally, with average errors of position being reduced by 97% and of orientation by 82%.
A related approach to improve the accuracy of industrial robots by considering their structural properties is presented in [60]. Factors such as manipulability, structural stiffness, inertia, damping, and natural frequencies are evaluated, and their effects on machining accuracy are studied. From experiments, the milling (end mill 6 mm in diameter) accuracy shows dependency on all considered properties, and tuning the robot’s posture improves the accuracy of a hole from 1.86 to 0.23 mm in diameter and from 0.87 to 0.16 mm in cylindricity. This improvement was achieved using the original machining accuracy prediction model, approximating the relationship between manipulability and accuracy, and a milling quality indicator according to the robot’s structural properties.
H. Hoai Nam et al. [61] focus on the multibody modeling aspects of an industrial robot for machining operations. Three model variations of gravity compensation, orthogonal compliance, and backlash are proposed to assess the effects of gravity, joint flexibility, and bending. The milling operations on aluminum and steel are modeled, and the sensitivity of each model is assessed. Gravity compensation is useful when the robotic structure is straightened horizontally. Orthogonal compliance is used to represent torsional stiffness and flexibilities, and backlash is used to describe the non-linear behavior to identify null torques. The most advanced model using backlash could accurately reproduce the cutting force signals and machined shapes in the test with aluminum but had some difficulties with steel. The authors conclude that a CAD-developed robot model in a machining context can suffice to estimate the cutting forces at moderate amplitudes.
In Ref. [62], a visual recognition system was proposed to recognize and obtain the positions of objects in the scene during manipulation tasks. The camera, mounted on the robot effector, causes a significant change in its position relative to the location as the robot moves to adapt its path and correctly recognize objects. The study focuses on industrial applications where all objects manipulated are of the same material, and it is impossible to recognize them by texture or color. The proposed system was designed using a specific distributed client and server architecture with implementation to overcome recognition issues when objects can only be recognized by geometric shapes. This method could also be adapted to the machining task to localize workpieces.
A dynamic pose correction scheme was offered in [63] to improve industrial robot pose accuracy implementing dynamic pose measurements as feedback (Figure 6). The optical coordinate measuring machine was used to measure the pose online, and a root mean square method was proposed for noise filtering from the measurements. The correction scheme adopted a PID controller and generated commands for the FANUC robot controller. The developed scheme was tested on two different six-DOF robots. A dual-camera sensor was used for online correction, and a dynamic pose correction (DPC) algorithm was suggested to improve the accuracy of an industrial robot for stationary tasks. Another implementation using the dynamic path modification, without the option, was tested, which applies dynamic corrections directly to the robot. Both implementations demonstrated improved positioning accuracy of 0.050 mm and 0.050° for orientation. Compared to the results of static calibration on one of the robots, the DPC greatly improved the pose accuracy, where the mean error was 0.092 mm and the maximum position error was 0.240 mm.
A robotic seam tracking system based on vision sensing and human–machine interaction based on the relationship between the images and the deviation of the welding torch was proposed and tested in [64]. The system uses the groove edges as a reference to control the deviation and correct it. The deviation is measured by comparing the left or right groove edge alternately with the relative positioning ruler from the real-time images obtained by the camera. The system’s accuracy, response rate, and smoothness were verified, and the average positioning error was less than 0.5 mm. This method can also be applied to the milling tasks to monitor the milled grooves.
As can be seen from the performed analysis, there are two main application cases when additional sensing technologies complement traditional setups of robotic material machining. In the first case, sensors obtain individual robot parameters to build predictive models or adapt existing models to exact units and tasks. The second case is real-time TCP tracing to obtain feedback data and implement a closed-loop control method. Both approaches have their strengths and limitations. The first case does not require measuring TCP position continuously, but on the other hand, predictive/past data-based models can be absolutely accurate and reliable. The second case is more reliable and common in practice but is sometimes limited by available sensing technologies, especially in material machining, where a non-contact measurement method can be required to ensure micrometric-scale measurement accuracy and resolution in a large part of the robot workspace.

5.3. Machine Learning Based Adaptive Solutions

Industrial robotics with ML has already enabled adaptive and intelligent systems that can learn data, optimize processes, and make real-time decisions, leading to increased efficiency, productivity, and flexibility in robotic manufacturing operations. Some research and implementation cases of machine learning adaptations are presented in Table 10.
The study presented in [62] uses machine learning algorithms for 3D object recognition and the correction of the robot approach path, assuming that the object does not change during manipulation. Performed experiments revealed a relationship between the success rate and the complexity of the object shape. Simple objects with few faces, e.g., revolution surfaces, can cause ambiguity, which can be corrected by increasing the number of views selected from the database. There is also a relationship between NN and runtime, as many comparisons in the compliant process can slow the 3D recognition, causing a delay in the robotic handling tasks. The proposed system using sensor data and robot kinematics determines the object’s pose and estimates where the object is placed in the robot’s path during the next iteration.
In [65], a machine-vision-based pose estimation system was developed for robotic machining, with fiducial markers designed to guarantee detection from all camera angles. An optimization method was used to minimize the errors in determining the corner points of the markers in the object frame, reducing the errors in the pose estimation. Two approaches were used to improve the accuracy of the assessed pose: LSTM NN and sparse regression. Both approaches used data from the Levenberg–Marquardt-based algorithm and the torques provided by the joints, trained with a laser tracker. The methods were validated via robot machining a part of NAS 979 material and free-form milling. The proposed LSTM NN and sparse regression methods proved more effective than the EKF approach regarding accuracy and precision. Using the EKF, the proposed LSTM network, and the proposed sparse regression approaches, the absolute position errors of 5.47 mm, 2.9 mm, and 2.05 mm were attained on average for NAS 979 machining, respectively. Similarly, on average, values of 5.35 mm, 2.17 mm, and 0.86 mm for free-form machining were attained. The proposed sparse regression-based method provided better results than the LSTM-based approach.
The implementation of ML can increase positioning accuracy in robotics [66]. A methodology for an online deep Q-learning-based approach is presented, which experimentally analyzes predetermined robot properties and their impact on overall accuracy. The research shows that the robot’s warming-up time, load, speed of movement, and trajectory affect its dynamic behavior. Relationships between the robot’s operating time, load, and positioning accuracy were defined. The proposed ML-based compensation method decreases positioning errors at key trajectory points by more than 30%.
Although most ML applications focus on accuracy enhancement, other parameters can also be set as the optimization goal. An energy-efficient trajectory planning method for robot manipulators is proposed in [67] using an ML-based approach. The movements of industrial robots are digitalized in joint space, and the industrial robots follow a collection of designed trajectories while the measurements of energy consumption are obtained. An ML model is trained using these datasets, which are used to provide a fitness function to obtain a near-optimal or optimal trajectory. The proposed method is validated using a simplified case study, and its major contributions are the ability to capture differences in energy consumption throughout the lifecycle of industrial robots and global optimization from an energy-saving perspective.
The positioning of industrial robots in precise machining is essential, but traditional reliability analysis methods produce unreliable results because of the complexity of the nonlinear system [68]. An HLA for training a radial basis function NN was proposed in the article, which is used to more efficiently and accurately evaluate the positioning accuracy reliability of robots. The HLA integrates the clustering and orthogonal least squares learning algorithms. The proposed method was proven in effective processing of high-dimension problems with uncertain distributional parameter types through simulations of four examples. Reliability analysis of positioning accuracy was conducted and showed similar results to the Monte Carlo simulation method with improved computational efficiency.
Ref. [69] discusses the transformation of manual manufacturing to robotic machining, emphasizing the need for a reliable in-process monitoring system. The authors propose a vision-based technique using deep learning for automatic endpoint detection of weld seams for their removal implementing robotic belt grinding. They employ CNN architecture for semantic segmentation of weld seam removal states, training on four different weld seam states. The proposed method provides competing segmentation accuracy and allows the detection of changes in weld profile geometry independent of grinding parameters. In addition, it shows promise for the removal of welds from free-form surfaces where toolpath planning and imaging are optimized. However, the paper acknowledges a higher misclassification between weld seam states and the background due to the larger area covered by the background. Future work will aim to improve network performance in localization, thereby enhancing real-time decision making.
Summarizing the analyzed research, it is possible to note that the typical use case for ML-based trajectory planning and control is insufficient positioning accuracy or the necessity to optimize robot behavior in a long-term period according to a given parameter. The main uncertainty of implementing ML is the complicated relationship between the amount of training data, algorithm performance, impact, and reliability of achieved results. Typically, the larger amount of training data results in more reliable ML output, but on the other hand, it increases training and model response time, which can affect the efficiency of the robotic cell or all manufacturing processes. Therefore, a detailed analysis of each use case is necessary before ML implementation.

6. Discussion

This review indicates that the knowledge of existing methods and technological solutions in robotic machining faces dynamics problems. Issues such as the stiffness-to-accuracy ratio, accuracy improvement methods, process optimization, adaptive feedback methods, robotic control, sensing, and machine learning algorithms have been systematized:
  • Robotic machining cases, robotic tool path and posture planning, advanced robotic technologies such as control, sensing, and machine learning, and robotic material processing technologies were qualitatively evaluated.
  • Robotic machining operations such as milling, grinding, polishing, and drilling were the most analyzed, as they are the most common operations used in industrial robotics and face dynamics issues. In most of the analyzed literature, the most common objective in all operations is to stiffen the construction of a robotic arm. Some offer methods to plan the robot’s posture or orient the tool, while others optimize machining parameters using advanced control techniques and collected sensor data.
  • Advanced control techniques include path planning to determine the optimal tool paths for machining, adaptive control to adjust robot movements in real time based on sensory feedback, and force control to apply consistent force during machining. Additionally, virtual and augmented reality and digital twin models are paving the way to innovative and advanced control solutions.
  • In recent research cases, force and torque sensors are used mainly to track the performance of the industrial robot in terms of resulting vibrations due to the machining process. More advanced sensing tools, such as machine vision systems and laser scanners, are useful for tracking the tool trajectory deviations or scanning the blanket surface to predict the best tool path.
  • Combining sensor data from multiple sources, there is a possibility of conducting adaptive path planning by implementing ML approaches. ML tools enable the achievement of advanced processes such as robot path optimization by simulating various cutting strategies, cutting tool wear prediction based on cutting conditions or material properties, quality control by analyzing sensor data in real time to detect defects or deviations, and overall process optimization.
Figure 7 illustrates key methods to achieve better characteristics, such as posture, accuracy, path, or stability, presented in the reviewed literature. Some of the methods or approaches are used in several of these, such as different static or dynamic models, performance indexes, FRF, point clouds, and different machine learning approaches. These approaches often have a common feature and are versatile tools that can be applied in various industrial robotics, e.g., static and dynamic models can help us understand the behavior of complex robotic systems, such as kinematic structures.
Performance indexes can be used to help evaluate the effectiveness or efficiency of a robotic system or machining process, where FRF is specifically used in analyzing vibration, which also influences the performance. ML has become popular in many areas due to its versatility and ability to compute large datasets and make predictions or classifications, including in industrial robotics. The presented analysis of the references revealed that robot accuracy and path generation are the most challenging issues in the field of robotic machining. The authors hope that this paper will foster more research in the field of robotic machining and broaden the area of practical applications.

7. Conclusions

The current situation in the field of industrial robot implementation in the area of machining indicates an increase in robot employment cases. The performed analysis revealed a few interesting tendencies:
  • Firstly, there is an evident distribution between the provided solutions and robotic quality parameters;
  • Of the eighteen solutions classified, thirteen rely on only quality parameters, four are important to two parameters, and only one solution has high importance to three parameters;
  • The presence of three quality parameters in the ML approach highlights the importance of bringing new perspectives to future solutions in robotics.
As the design of industrial robots, including robot body materials, motors, and drives, reaches a certain level, future enhancement of low robot stiffness, as mentioned, is unlikely. It is the opinion of the authors that dynamic methods of robot error compensation, including wide implementation of AI and, in particular, machine learning, provide a broad perspective to widen the implementation of robotic machining in industry, especially in flexible low-production-scale companies.

Author Contributions

Conceptualization, A.D. and V.B.; methodology, S.P.; validation, S.P., J.R. and A.D.; formal analysis, V.B.; investigation, M.M.; writing—original draft preparation, M.M.; writing—review and editing, J.R., A.D. and V.B.; visualization, S.P.; supervision, A.D.; funding acquisition, V.B. All authors have read and agreed to the published version of the manuscript.


This project has received financial support from the Research Council of Lithuania (LMTLT), Nr. P-LLT-21-6, State Education Development Agency of Latvia, Ministry of Science and Technology (MOST)) of Taiwan.

Data Availability Statement

During this research, no new data were created.

Conflicts of Interest

The authors declare no conflict of interest.


  1. Lv, Y.; Peng, Z.; Qu, C.; Zhu, D. An Adaptive Trajectory Planning Algorithm for Robotic Belt Grinding of Blade Leading and Trailing Edges Based on Material Removal Profile Model. Robot. Comput. Integr. Manuf. 2020, 66, 101987. [Google Scholar] [CrossRef]
  2. Lu, L.; Zhang, J.; Fuh, J.Y.H.; Han, J.; Wang, H. Time-Optimal Tool Motion Planning with Tool-Tip Kinematic Constraints for Robotic Machining of Sculptured Surfaces. Robot. Comput. Integr. Manuf. 2020, 65, 101969. [Google Scholar] [CrossRef]
  3. Haage, M.; Profanter, S.; Kessler, I.; Somani, N.; Karlsson, M.; Robertz, S.G.; Marti, M. On Cognitive Robot Woodworking in SMErobotics. In Proceedings of the ISR 2016: 47st International Symposium on Robotics, Munich, Germany, 21–22 June 2016; VDE: Munich, Germany, 2016. [Google Scholar]
  4. Quintero, C.P.; Li, S.; Pan, M.K.; Chan, W.P.; Machiel Van Der Loos, H.F.; Croft, E. Robot Programming Through Augmented Trajectories in Augmented Reality. In Proceedings of the 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Madrid, Spain, 1–5 October 2018; pp. 1838–1844. [Google Scholar] [CrossRef]
  5. Kubela, T.; Pochyly, A.; Singule, V. Assessment of Industrial Robots Accuracy in Relation to Accuracy Improvement in Machining Processes. In Proceedings of the 2016 IEEE International Power Electronics and Motion Control Conference (PEMC), Varna, Bulgaria, 25–28 September 2016; pp. 720–725. [Google Scholar] [CrossRef]
  6. Schneider, U.; Drust, M.; Ansaloni, M.; Lehmann, C.; Pellicciari, M.; Leali, F.; Gunnink, J.W.; Verl, A. Improving Robotic Machining Accuracy through Experimental Error Investigation and Modular Compensation. Int. J. Adv. Manuf. Technol. 2016, 85, 3–15. [Google Scholar] [CrossRef]
  7. Ji, W.; Wang, L. Industrial Robotic Machining: A Review. Int. J. Adv. Manuf. Technol. 2019, 103, 1239–1255. [Google Scholar] [CrossRef]
  8. Hyeon Kim, S.; Nam, E.; In Ha, T.; Hwang, S.-H.; Ho Lee, J.; Park, S.-H.; Min, B.-K. Robotic Machining: A Review of Recent Progress. Int. J. Precis. Eng. Manuf. 2019, 20, 1629–1642. [Google Scholar] [CrossRef]
  9. Iglesias, I.; Sebastián, M.A.; Ares, J.E. Overview of the State of Robotic Machining: Current Situation and Future Potential. Procedia Eng. 2015, 132, 911–917. [Google Scholar] [CrossRef]
  10. Verl, A.; Valente, A.; Melkote, S.; Brecher, C.; Ozturk, E.; Tunc, L.T. Robots in Machining. CIRP Ann. 2019, 68, 799–822. [Google Scholar] [CrossRef]
  11. Cvitanic, T.; Nguyen, V.; Melkote, S.N. Pose Optimization in Robotic Machining Using Static and Dynamic Stiffness Models. Robot. Comput. Integr. Manuf. 2020, 66, 101992. [Google Scholar] [CrossRef]
  12. Kaldestad, K.B.; Tyapin, I.; Hovland, G. Robotic Face Milling Path Correction and Vibration Reduction. In Proceedings of the IEEE International Conference on Advanced Intelligent Mechatronics (AIM), Busan, Korea, 7–11 July 2015; pp. 543–548. [Google Scholar] [CrossRef]
  13. Karim, A.; Hitzer, J.; Lechler, A.; Verl, A. Analysis of the Dynamic Behavior of a Six-Axis Industrial Robot within the Entire Workspace in Respect of Machining Tasks. In Proceedings of the 2015 IEEE International Conference on Advanced Intelligent Mechatronics (AIM), Busan, Republic of Korea, 7–11 July 2017; pp. 670–675. [Google Scholar] [CrossRef]
  14. Sun, Y.; Shi, Z.; Guo, Q.; Xu, J. A Novel Method to Predict Surface Topography in Robotic Milling of Directional Plexiglas Considering Cutter Dynamical Displacement. J. Mater. Process. Technol. 2022, 304, 117545. [Google Scholar] [CrossRef]
  15. Pérez-Ruiz, J.D.; de Lacalle, L.N.L.; Urbikain, G.; Pereira, O.; Martínez, S.; Bris, J. On the Relationship between Cutting Forces and Anisotropy Features in the Milling of LPBF Inconel 718 for near Net Shape Parts. Int. J. Mach. Tools Manuf. 2021, 170, 103801. [Google Scholar] [CrossRef]
  16. Gonzalez, M.K.; Theissen, N.A.; Barrios, A.; Archenti, A. Online Compliance Error Compensation System for Industrial Manipulators in Contact Applications. Robot. Comput. Integr. Manuf. 2022, 76, 102305. [Google Scholar] [CrossRef]
  17. Chen, C.; Peng, F.; Yan, R.; Fan, Z.; Li, Y.; Wei, D. Posture-Dependent Stability Prediction of a Milling Industrial Robot Based on Inverse Distance Weighted Method. Procedia Manuf. 2018, 17, 993–1000. [Google Scholar] [CrossRef]
  18. Nguyen, V.; Johnson, J.; Melkote, S. Active Vibration Suppression in Robotic Milling Using Optimal Control. Int. J. Mach. Tools Manuf. 2020, 152, 103541. [Google Scholar] [CrossRef]
  19. Chen, C.; Peng, F.; Yan, R.; Li, Y.; Wei, D.; Fan, Z.; Tang, X.; Zhu, Z. Stiffness Performance Index Based Posture and Feed Orientation Optimization in Robotic Milling Process. Robot. Comput. Integr. Manuf. 2019, 55, 29–40. [Google Scholar] [CrossRef]
  20. Xiong, G.; Ding, Y.; Zhu, L.M. Stiffness-Based Pose Optimization of an Industrial Robot for Five-Axis Milling. Robot. Comput. Integr. Manuf. 2019, 55, 19–28. [Google Scholar] [CrossRef]
  21. Wang, Z.Q.; Liu, X.Q.; Wang, X.R.; Li, C.Y.; Yang, N.; Lin, T.S. Robotic Milling of Complex NURBS Surface with Fixed Cutter Axis Control Method. Ind. Robot. 2021, 48. [Google Scholar] [CrossRef]
  22. Zhu, D.; Feng, X.; Xu, X.; Yang, Z.; Li, W.; Yan, S.; Ding, H. Robotic Grinding of Complex Components: A Step towards Efficient and Intelligent Machining—Challenges, Solutions, and Applications. Robot. Comput. Integr. Manuf. 2020, 65, 101908. [Google Scholar] [CrossRef]
  23. Yin, F.; Wu, S.; Huang, H.; Cui, C.; Ji, Q. Effect of Machining Trajectory on Grinding Force of Complex-Shaped Stone by Robotic Manipulator. Machines 2022, 10, 787. [Google Scholar] [CrossRef]
  24. Xie, H.; Li, W.L.; Zhu, D.H.; Yin, Z.P.; Ding, H. A Systematic Model of Machining Error Reduction in Robotic Grinding. IEEE/ASME Trans. Mechatron. 2020, 25, 2961–2972. [Google Scholar] [CrossRef]
  25. Zhou, P.; Zhao, X.; Tao, B.; Ding, H. Time-Varying Isobaric Surface Reconstruction and Path Planning for Robotic Grinding of Weak-Stiffness Workpieces. Robot. Comput. Integr. Manuf. 2020, 64, 101945. [Google Scholar] [CrossRef]
  26. Wan, G.; Wang, G.; Fan, Y. A Robotic Grinding Station Based on an Industrial Manipulator and Vision System. PLoS ONE 2021, 16, e0248993. [Google Scholar] [CrossRef]
  27. Xie, Y.; Chang, G.; Yang, J.; Zhao, M.; Li, J. Process Optimization of Robotic Polishing for Mold Steel Based on Response Surface Method. Materials 2022, 10, 283. [Google Scholar] [CrossRef]
  28. Liang, X.; Cai, Z.; Zeng, C.; Mu, Z.; Li, Z.; Yang, F.; Chen, T.; Dong, S.; Deng, C.; Niu, S. A Robotic Polishing Trajectory Planning Method for TBCs of Aero-Engine Turbine Blade Using Measured Point Cloud. Ind. Robot. 2023, 50, 275–286. [Google Scholar] [CrossRef]
  29. Zhu, R.; Yang, G.; Fang, Z.; Dai, J.; Chen, C.Y.; Zhang, G.; Zhang, C. Hybrid Orientation/Force Control for Robotic Polishing with a 2R1T Force-Controlled End-Effector. Int. J. Adv. Manuf. Technol. 2022, 121, 2279–2290. [Google Scholar] [CrossRef]
  30. Wahballa, H.; Duan, J.; Dai, Z. Constant Force Tracking Using Online Stiffness and Reverse Damping Force of Variable Impedance Controller for Robotic Polishing. Int. J. Adv. Manuf. Technol. 2022, 121, 5855–5872. [Google Scholar] [CrossRef]
  31. Zhang, J.; Zhao, L.; Li, L.; Ma, F.; Chen, G. Design of Passive Constant-Force End-Effector for Robotic Polishing of Optical Reflective Mirrors. Chin. J. Mech. Eng. 2022, 35, 141. [Google Scholar] [CrossRef]
  32. Wei, Y.; Xu, Q. Design of a New Passive End-Effector Based on Constant-Force Mechanism for Robotic Polishing. Robot. Comput. Integr. Manuf. 2022, 74, 102278. [Google Scholar] [CrossRef]
  33. Tian, F.; Li, Z.; Lv, C.; Liu, G. Polishing Pressure Investigations of Robot Automatic Polishing on Curved Surfaces. Int. J. Adv. Manuf. Technol. 2016, 87, 639–646. [Google Scholar] [CrossRef]
  34. Wang, Z.; Zhang, R.; Keogh, P. Real-Time Laser Tracker Compensation of Robotic Drilling and Machining. J. Manuf. Mater. Process. 2020, 4, 79. [Google Scholar] [CrossRef]
  35. Lu, H.; Zhao, X.; Tao, B.; Ding, H. A State-Classification Approach for Light-Weight Robotic Drilling Using Model-Based Data Augmentation and Multi-Level Deep Learning. Mech. Syst. Signal Process. 2022, 167, 108480. [Google Scholar] [CrossRef]
  36. Zhang, J.; Liao, W.; Bu, Y.; Tian, W.; Hu, J. Stiffness Properties Analysis and Enhancement in Robotic Drilling Application. Int. J. Adv. Manuf. Technol. 2020, 106, 5539–5558. [Google Scholar] [CrossRef]
  37. Arthur, J.; Khoshdarregi, M. Pose Optimization and Path Improvement in Robotic Drilling through Minimization of Joint Reversals. Adv. Robot. 2022, 2022, 1076–1086. [Google Scholar] [CrossRef]
  38. Pereira, B.; Griffiths, C.A.; Birch, B.; Rees, A. Optimization of an Autonomous Robotic Drilling System for the Machining of Aluminum Aerospace Alloys. Int. J. Adv. Manuf. Technol. 2022, 119, 2429–2444. [Google Scholar] [CrossRef]
  39. Farhadi, A.; Lee, S.K.H.; Hinchy, E.P.; O’dowd, N.P.; Avizzano, C.A.; Farhadi, A.; Lee, S.K.H.; Hinchy, E.P.; O’dowd, N.P.; Mccarthy, C.T. The Development of a Digital Twin Framework for an Industrial Robotic Drilling Process. Sensors 2022, 22, 7232. [Google Scholar] [CrossRef]
  40. Onstein, I.F.; Semeniuta, O.; Bjerkeng, M. Deburring Using Robot Manipulators: A Review. In Proceedings of the 2020 3rd International Symposium on Small-Scale Intelligent Manufacturing Systems (SIMS 2020), Gjøvik, Norway, 10–12 June 2020. [Google Scholar] [CrossRef]
  41. Gonzalez, M.; Rodriguez, A.; Pereira, O.; Celaya, A.; Lopez de Lacalle, L.N.; Esparta, M. Axial-Compliant Tools for Adaptive Chamfering of Sharp-Edges: Characterisation and Modelling. Eng. Sci. Technol. Int. J. 2023, 41, 101407. [Google Scholar] [CrossRef]
  42. Hu, J.; Pagilla, P.R. A Novel Force and Motion Control Strategy for Robotic Chamfering of Gears. IFAC-PapersOnLine 2020, 53, 8710–8715. [Google Scholar] [CrossRef]
  43. Miller, G.; Irani, R.A.; Ahmadi, M. The Application of Mechanistic Cutting Force Models for Robotic Deburring. Int. J. Adv. Manuf. Technol. 2021, 115, 199–212. [Google Scholar] [CrossRef]
  44. Wang, G.; Dong, H.; Guo, Y.; Ke, Y. Early Chatter Identification of Robotic Boring Process Using Measured Force of Dynamometer. Int. J. Adv. Manuf. Technol. 2018, 94, 1243–1252. [Google Scholar] [CrossRef]
  45. Lin, Y.; Zhao, H.; Ding, H. Real-Time Path Correction of Industrial Robots in Machining of Large-Scale Components Based on Model and Data Hybrid Drive. Robot. Comput. Integr. Manuf. 2023, 79, 102447. [Google Scholar] [CrossRef]
  46. Cordes, M.; Hintze, W. Offline Simulation of Path Deviation Due to Joint Compliance and Hysteresis for Robot Machining. Int. J. Adv. Manuf. Technol. 2017, 90, 1075–1083. [Google Scholar] [CrossRef]
  47. Stepputat, M.; Beuss, F.; Pfletscher, U.; Sender, J.; Fluegge, W. Automated One-off Production in Woodworking by Part-to-Tool. Procedia CIRP 2021, 104, 307–312. [Google Scholar] [CrossRef]
  48. Zhou, J.; Cao, H.; Jiang, P.; Li, C.; Yi, H.; Liu, M. Energy-Saving Trajectory Planning for Robotic High-Speed Milling of Sculptured Surfaces. IEEE Trans. Autom. Sci. Eng. 2022, 19, 2278–2294. [Google Scholar] [CrossRef]
  49. Munasinghe, N.; Paul, G. Path Planning for Robot Based Radial Advanced Manufacturing Using Print Space Sampling. In Proceedings of the 16th IEEE International Conference on Control, Automation, Robotics and Vision (ICARCV), Shenzhen, China, 13–15 December 2020; pp. 854–859. [Google Scholar] [CrossRef]
  50. Wang, G.; Li, W.; Jiang, C.; Zhu, D.; Li, Z.; Xu, W.; Zhao, H.; Ding, H. Trajectory Planning and Optimization for Robotic Machining Based on Measured Point Cloud. IEEE Trans. Robot. 2022, 38, 1621–1637. [Google Scholar] [CrossRef]
  51. Sánchez, I.I.; Ares, J.E.; Gaya, C.G.; Prieto, V.R. A New Approach to the Consideration and Analysis of Critical Factors in Robotic Machining. Appl. Sci. 2020, 10, 8885. [Google Scholar] [CrossRef]
  52. Sun, Y.; Jia, J.; Xu, J.; Chen, M.; Niu, J. Path, Feedrate and Trajectory Planning for Free-Form Surface Machining: A State-of-the-Art Review. Chin. J. Aeronaut. 2022, 35, 12–29. [Google Scholar] [CrossRef]
  53. Morozov, M.; Pierce, S.G.; MacLeod, C.N.; Mineo, C.; Summan, R. Off-Line Scan Path Planning for Robotic NDT. Measurement 2018, 122, 284–290. [Google Scholar] [CrossRef]
  54. Pham, Q.C.; Stasse, O. Time-Optimal Path Parameterization for Redundantly Actuated Robots: A Numerical Integration Approach. IEEE/ASME Trans. Mechatron. 2015, 20, 3257–3263. [Google Scholar] [CrossRef]
  55. Lin, Y.; Zhao, H.; Ding, H. Posture Optimization Methodology of 6R Industrial Robots for Machining Using Performance Evaluation Indexes. Robot. Comput. Integr. Manuf. 2017, 48, 59–72. [Google Scholar] [CrossRef]
  56. Hu, J.; Li, C.; Chen, Z.; Yao, B. Precision Motion Control of a 6-DoFs Industrial Robot with Accurate Payload Estimation. IEEE/ASME Trans. Mechatron. 2020, 25, 1821–1829. [Google Scholar] [CrossRef]
  57. Wu, K.; Krewet, C.; Kuhlenkötter, B. Dynamic Performance of Industrial Robot in Corner Path with CNC Controller. Robot. Comput. Integr. Manuf. 2018, 54, 156–161. [Google Scholar] [CrossRef]
  58. Moeller, C.; Schmidt, H.C.; Koch, P.; Boehlmann, C.; Kothe, S.; Wollnack, J.; Hintze, W.; Moeller, C.; Schmidt, C.; Koch, H.; et al. Real Time Pose Control of an Industrial Robotic System for Machining of Large Scale Components in Aerospace Industry Using Laser Tracker System “Real Time Pose Control of an Industrial Robotic System for Machining of Large Scale Components in Aerospace Industry Using Laser Tracker System. SAE Int. J. Aerosp. 2017, 10, 100–108. [Google Scholar] [CrossRef]
  59. Yamaguchi, N.; Aiuchi, K.; Morita, N.; Miyashita, J.; Moriya, M.; Maeda, K.; Uchiyama, N. Calibration of Robotic Wood working Machinery Using a Motion Capture System. In Proceedings of the 2022 IEEE/SICE International Symposium on System Integration (SII), Narvik, Norway, 9–12 January 2022. [Google Scholar] [CrossRef]
  60. Janez, G.; Timi, K.; Karl, G.; Miran, B. Accuracy Improvement of Robotic Machining Based on Robot’s Structural Properties. Int. J. Adv. Manuf. Technol. 2020, 108, 1309–1329. [Google Scholar] [CrossRef]
  61. Hoai Nam, H.; Riviere, E.; Nam Huynh, H.; Ere-Lorphèvre, E.; Verlinden, O. Multibody Modelling of a Flexible 6-Axis Robot Dedicated to Robotic Machining. In Proceedings of the 5th Joint International Conference on Multibody System Dynamics (IMSD), Lisbon, Portugal, 24–28 June 2018. [Google Scholar]
  62. Mateo, C.M.; Gil, P.; Torres, F. Visual Perception for the 3D Recognition of Geometric Pieces in Robotic Manipulation. Int. J. Adv. Manuf. Technol. 2016, 83, 1999–2013. [Google Scholar] [CrossRef]
  63. Gharaaty, S.; Shu, T.; Joubair, A.; Xie, W.F.; Bonev, I.A. Online Pose Correction of an Industrial Robot Using an Optical Coordinate Measure Machine System. Sage J. 2018, 15, 1–16. [Google Scholar] [CrossRef]
  64. Xue, K.; Wang, Z.; Shen, J.; Hu, S.; Zhen, Y.; Liu, J.; Wu, D.; Yang, H. Robotic Seam Tracking System Based on Vision Sensing and Human-Machine Interaction for Multi-Pass MAG Welding. J. Manuf. Process. 2021, 63, 48–59. [Google Scholar] [CrossRef]
  65. Bilal, D.K.; Unel, M.; Tunc, L.T.; Gonul, B. Development of a Vision Based Pose Estimation System for Robotic Machining and Improving Its Accuracy Using LSTM Neural Networks and Sparse Regression. Robot. Comput. Integr. Manuf. 2022, 74, 102262. [Google Scholar] [CrossRef]
  66. Bucinskas, V.; Dzedzickis, A.; Sumanas, M.; Sutinys, E.; Petkevicius, S.; Butkiene, J.; Virzonis, D.; Morkvenaite-Vilkonciene, I. Improving Industrial Robot Positioning Accuracy to the Microscale Using Machine Learning Method. Machines 2022, 10, 940. [Google Scholar] [CrossRef]
  67. Yin, S.; Ji, W.; Wang, L. A Machine Learning Based Energy Efficient Trajectory Planning Approach for Industrial Robots. Procedia CIRP 2019, 81, 429–434. [Google Scholar] [CrossRef]
  68. Zhang, D.; Zhang, N.; Ye, N.; Fang, J.; Han, X. Hybrid Learning Algorithm of Radial Basis Function Networks for Reliability Analysis. IEEE Trans. Reliab. 2021, 70, 887–900. [Google Scholar] [CrossRef]
  69. Pandiyan, V.; Murugan, P.; Tjahjowidodo, T.; Caesarendra, W.; Manyar, O.M.; Then, D.J.H. In-Process Virtual Verification of Weld Seam Removal in Robotic Abrasive Belt Grinding Process Using Deep Learning. Robot. Comput. Integr. Manuf. 2019, 57, 477–487. [Google Scholar] [CrossRef]
Figure 1. The example representation of industrial robotic machining.
Figure 1. The example representation of industrial robotic machining.
Robotics 12 00160 g001
Figure 2. Scheme of the process for article inclusion in the systematic review.
Figure 2. Scheme of the process for article inclusion in the systematic review.
Robotics 12 00160 g002
Figure 3. Summary of machining processes.
Figure 3. Summary of machining processes.
Robotics 12 00160 g003
Figure 4. A framework of a real-time path correction approach based on JPE [45].
Figure 4. A framework of a real-time path correction approach based on JPE [45].
Robotics 12 00160 g004
Figure 5. Impact test setup in [17].
Figure 5. Impact test setup in [17].
Robotics 12 00160 g005
Figure 6. Robotic sensing case using the dynamic pose correction algorithm with visual sensor [63].
Figure 6. Robotic sensing case using the dynamic pose correction algorithm with visual sensor [63].
Robotics 12 00160 g006
Figure 7. Methods used in the reviewed literature.
Figure 7. Methods used in the reviewed literature.
Robotics 12 00160 g007
Table 2. Summary of robotic grinding.
Table 2. Summary of robotic grinding.
ProblemMethods Industrial RobotAdditional MeansRef.
Poor grinding accuracy and surface qualityReview on calibration and measurement, trajectory planning, force control, and surface integrity-Active/passive force
control sensors,
PID-based control
Examination of curvature characteristics of complex-shaped stone productsThe matching relationship between surface characteristics and machining trajectoryKUKA QUANTECK KR240 R2900 ultraKistler’s 9170B 6-dimensional torque sensor.[23]
Free-form surface machining is limited by individual kinematic errors or joint stiffness.Speed and force adjoint transformation, fine-tuning the workpiece frame positionABB industrial robot with RobotStudio 2020.3 softwareGrinding machine with a belt, robot here is a
workpiece manipulator
Profile accuracy enhancementTrajectory planning algorithm adapting the material removal profile (MRP)-Trajectory planning
software based on
Grinding of weak-stiffness workpiecesDeformation and stiffness measurements and time-varying isobaric surface (TVIS) mesh generationUniversal Robot UR5Six-dimensional force sensor ATI Axia80 [25]
Increasing demands for precision and automationGeometric-multilevel-Line-2D methodEFORT ER50-C10,
Different shape parts,
vision system,
PC workstation
(CPU: i7, GPU: RTX 2070S, RAM: 32G)
Table 3. Summary of robotic polishing.
Table 3. Summary of robotic polishing.
ProblemMethods Industrial RobotAdditional MeansRef.
Surface roughness reduction and improvement of surface quality of mold steelOptimizations of process parameters: polishing pressure, feed speed, and rotating speed of the toolKUKA KR60-3Force control with a six-dimensional force sensor[27]
Surface roughness reduction of bladesPoint cloud preprocessing, slicing algorithm, and the
intersection method
1200–7/0.7 with
RobotStudio simulations
3D profile sensor[28]
Achievement of consistent surface qualityContact point compensation model to predict the contact point variationABB industrial robotForce-controlled
Desire constant force tracking controlImpedance controller with online stiffness and reverse damping force7-DOF X-mate3- ProLyapunov function,
force sensor
Maintenance of a constant force between the robot and the workpieceConstant force compliant mechanism, compared to traditional methods using force control7-DOF KUKA industrial robotPassive constant force end-effector,
K9 R467 reflective mirror
Achievement of constant
polishing pressure
Preston equation and Hertz theory-based constant force mechanism Universal Robot UR5Constant force
Improvement of finishing
efficiency, surface quality, and surface consistency
A relation model between
removal rate and polishing pressure
Kuka KR30-3Force/torque sensor[33]
Table 4. Summary of robotic drilling.
Table 4. Summary of robotic drilling.
ProblemMethods Industrial RobotAdditional MeansRef.
Low structural stiffness and low positional accuracyBallbar dynamic path accuracy, a series of drilling case studies, and machining testsKUKA KR120R2500 PRO with a KUKA KR C4 controllerSingle 3-DOF laser tracker[34]
Detection of unqualified holes caused by inclined drillingVibration-based classificationUR industrial robotResnet classifier with
vibration model, camera
Robot stiffness influence on drilling qualityPreload pressing force to strengthen the stiffness of the machining planeKUKA industrial robotZEISS SPECTRUM II Coordinate measuring machine[36]
Static friction in robot joints impacts the qualityOptimization framework, which models a general drilling motion minimizing joint reversalsKUKA KR 6 R700-2Particle Swarm Optimization[37]
Hole surface roughness and exit burr heightsTaguchi design methodologyKuka KR16CNC milling machine as a reference, SignalCalc Ace Vibration sensors[38]
A unified digital twin framework for the manufacturing environment is missing.Generic reference model to highlight elements of the digital twin KUKA KR210 R3100 Ultra with a KRC4 controllerBased on ISO 23247 standard, the drilling and vision module [39]
Table 5. Summary of other machining cases in robotics.
Table 5. Summary of other machining cases in robotics.
ProblemMethods Industrial RobotAdditional MeansRef.
Poor positioning accuracyForce controlABB IRB-4600Cone-shape stone tool, Omega 85 force-torque sensor, AFD 310 compliant device, spindle,
robot operating system (ROS)
The need for suitable force
Linear regression method,
exponential model based on the simplex search method
Unspecified six-DOF robotForce-torque sensor, laser scanner, spindle, aluminum 6061 workpiece[43]
Poor quality and efficiency of the hole Prediction based on measured force, intrinsic mode functions, empirical mode decomposition, Hilbert transform, and spectrum ABB-IRB6600 Kistler9257B dynamometer, end-effector, mobile platform, control system[44]
Table 6. Summary of robotic path planning.
Table 6. Summary of robotic path planning.
ProblemMethods Industrial RobotAdditional MeansRef.
Insufficient manipulator
Real-time path correction
applying joint
position error (JPE)
ODG-JLRB20flexible-dynamics-based disturbance Kalman filter (FDBDKF), force/torque sensor, MATLAB[45]
Path deviation predictionBased on joint stiffness and reversal errorABB IRB 6660-205/1.9Laser tracker,
dynamometer, reflector
Insufficient manipulator
Automatically Programmed-Tool (APT) code to generate Part-to-Tool (PtT) pathNot specified six-DOF
industrial robot
Severe deformations and
Deformation and stiffness measurements, and TVIS mesh generationUniversal Robot UR5Six-dimensional force sensor ATI Axia80 [25]
The influence of the trajectory planning method resulting poor accuracy of blade edgesIso-scallop height algorithm, material removal profile -A 6-axis force/torque sensor, trajectory
planning software based on OpenCASCADE
Minimization of energyEnergy characteristic to
acquire the energy-optimal feed rate
KUKA KR60-3Computer (Intel i3-8100 CPU 3.60 GHz and 8-GB DDR2)[48]
Distance error optimization and manipulabilityCost-based path planning to adapt print-space
ABB IRB 120MATLAB robotics toolbox[49]
Machining of large and
complex parts
Energy-based trajectory smoothness optimization method based on the point cloudsABB IRB 6660-205/1.9 Laser tracker, T-scan 3D scanner [50]
Identification of critical
factors affecting the
machining path
Predictive methodology ABB IRB 6660-205/1.9MITUTOYO coordinate
measuring machine
Tool path generation, feed rate scheduling, and trajectory planningCuter contact path
construction methods
ABB industrial robotThe 3/5-axis CNC machine tools (for reference), laser sensor, rotary dynamometer[52]
Curved tool path complexity and time-optimal motion
Pontryagin maximum
Elfin 5 with ROSMoveit library[2]
Computer-aided scan path generationSwept frequency eddy
currents method
KUKA KR5 with
MasterCAM X6
Laser tracker [53]
Table 7. Summary of robotic posture planning.
Table 7. Summary of robotic posture planning.
ProblemMethods Industrial RobotAdditional MeansRef.
Minimization of end-effector displacementStatic/dynamic model-based pose selectionKUKA KR 500–3Six-DOF laser tracker,
six-axis force/torque sensor
Prediction of the industrial
robot stability at any posture
FRF at the tool tipABB IRB 6660-205Sound microphone, NI data acquisition system[17]
Analyzation of the stiffness propertiesTask-dependent PIABB IRB 6660-205NI data acquisition system, laser tracker,
Posture optimizationBased on robotic PIs and stiffness mapComau Smart5 NJ 220-2.7Laser tracker and
Table 8. Summary of robotic control.
Table 8. Summary of robotic control.
ProblemControl Methods Industrial RobotAdditional MeansRef.
Simplification of the
excitation trajectory
DIARC algorithmCOMAU-RACER3MATLAB, Automation studio IDE, encoders[56]
Improvement of the
machining accuracy
Use of a CNC controller as the control systemKUKA KR 210 R2700High-speed camera, KR C4 controller [57]
Trajectory interaction and ease the robot programmingAugmented reality (AR) robotic control systemBarrett Whole-Arm
7-DOF Manipulator
Head-mounted display (Microsoft Hololens), MYO armband[4]
Table 9. Summary of robotic sensing.
Table 9. Summary of robotic sensing.
ProblemSensors Industrial RobotAdditional MeansRef.
Poor absolute accuracyLaser tracker systemMABI Robotic Max 150Siemens Sinumerik 840D sl-CNC controller,
Expensive calibrationMotion capture systemFANUC industrial robotMATLAB Optimization Toolbox, ROBOGUIDE Simulator[59]
Poor milling accuracyBody mounted accelerometerACMA XR701 Impact hammer tests and FEM model calibration[60]
The effects assessment of gravity, joint flexibility, and bendingForce sensorStabil TX200The use of torsional springs and dampers[61]
Obtain the positions of objects during grasping and robotic manipulationVisual recognition systemMitsubishi PA-10 7-DOF Client-server
architecture and communication via ARCNET
Poor pose accuracyOptical coordinate measuring machineFANUC LR Mate 200iC and FANUC M20iARoot mean square method, dynamic pose correction algorithm[63]
Control of the deviations and correctionVision sensing (CCD and
KUKA KR 16 L8 with KUKA robot controller (KRC)Human-machine
interaction module,
programmable logic
Table 10. Summary of ML applications in robotics.
Table 10. Summary of ML applications in robotics.
ApplicationML ModelsIndustrial RobotAdditional MeansRef.
3D object recognitionNeural Network (NN)Mitsubishi PA-10Microsoft Kinect™ RGBD range sensor[62]
Monocular machine vision-based pose estimationLong Short-Term Memory (LSTM) NNKUKA KR240 R2900Extended Kalman Filter (EKF) (for comparison) [65]
Analyzation of experimentally predetermined robot properties and their impact on overall accuracyDeep Q-learning-based
KUKA-YouBotTwo USB cameras,
accelerometers, data
acquisition system
Energy-efficient trajectory planningDeep learning network, evolutional-based or swarm-intelligence-based
Not specified six-DOF
industrial robot
Positioning accuracy
reliability analysis
Hybrid learning algorithm (HLA) for training a radial
basis function network
Not specified six-DOF
industrial robot
Monte Carlo simulation method (for comparison), MATLAB [68]
Weld seam removal with robotic grinding processConvolutional neural network (CNN) architectureABB 6660-205-19Camera system with
resolution of
1240 × 960 pixels, MATLAB
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Makulavičius, M.; Petkevičius, S.; Rožėnė, J.; Dzedzickis, A.; Bučinskas, V. Industrial Robots in Mechanical Machining: Perspectives and Limitations. Robotics 2023, 12, 160.

AMA Style

Makulavičius M, Petkevičius S, Rožėnė J, Dzedzickis A, Bučinskas V. Industrial Robots in Mechanical Machining: Perspectives and Limitations. Robotics. 2023; 12(6):160.

Chicago/Turabian Style

Makulavičius, Mantas, Sigitas Petkevičius, Justė Rožėnė, Andrius Dzedzickis, and Vytautas Bučinskas. 2023. "Industrial Robots in Mechanical Machining: Perspectives and Limitations" Robotics 12, no. 6: 160.

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop