Robot control system based on deep learning and RPA

Abstract

Robots can help people complete repetitive and high-risk tasks, such as industrial production, medical care, environmental monitoring, etc. The control system of robots is the key to their ability to complete tasks, and studying robot control systems is of great significance. This article used Convolutional Neural Network (CNN) and Robotic Process Automation (RPA) technologies to optimize and train the robot control system and constructed a robot control system. This article conducts perception and decision-making experiments and execution experiments in the experimental section. According to the experimental results, it can be concluded that the average image recognition accuracy of the robot control system in perception and decision-making experiments was 94.62%. The average decision accuracy was 87.5%, and the average time efficiency was 176 seconds. During the execution of the experiment, the deviation of the motion trajectory shall not exceed 5 cm, and the oscillation amplitude shall not exceed 6°; the distance from the obstacle shall not exceed 20 cm, and the movement speed shall not exceed 0.6 m/s; the operating time shall not exceed 25 hours, and the number of faults shall not exceed 0.2 times per hour, all within the normal range. The robot control system based on Deep Learning (DL) and RPA has broad application prospects and research value, which would bring new opportunities and challenges to the development and application of robot technology.

Keywords

Robot control system Robotic Process Automation (RPA)Convolutional Neural Network (CNN)Deep Learning (DL)

1 Introduction

With the continuous development and popularization of artificial intelligence technology, robots have become an indispensable part. The robot control system based on DL and RPA is a new type of robot control system that can control robots through DL algorithms and RPA technology. Compared with traditional robot control systems, robot control systems based on DL and RPA have the following advantages: DL algorithms can improve the recognition and judgment abilities of robots by learning a large amount of data, thereby achieving precise control of robots. RPA technology can enable robots to automate tasks and improve their work efficiency. DL algorithms can enable robots to have more intelligent functions, such as speech recognition and image recognition. RPA technology can enable robots to perform tasks autonomously without human intervention, thereby improving their intelligence level. DL algorithms can enable robots to more accurately recognize the surrounding environment and obstacles, thereby avoiding collisions between robots and the surrounding environment and personnel. RPA technology can make robots perform tasks more accurately and safely, improving the reliability of robots. The robot control system based on DL and RPA has important background significance. It can improve the efficiency, intelligence, safety, and reliability of robots, providing a broader space for their application in various fields. At the same time, it also provides strong support for the development and popularization of artificial intelligence technology.

With the continuous development of robot technology, robot control systems would also be continuously upgraded and optimized, providing more accurate, efficient, intelligent, and safe control methods for robot applications. Shared control is an increasingly popular method for promoting control and communication between humans and intelligent machines. However, there is almost no consensus on the design and evaluation guidelines for shared controls, and even on the definition of the composition of shared controls. The lack of consensus makes the cross-fusion of shared control research among different application fields complex. Abbink David A’s research reveals the generalizability and practicality of the proposed shared control framework in designing useful, safe, and comfortable interactions between humans and intelligent machines [1]. Wang Yueying proposed a reliable intelligent path following control method for airship robots with sensor failures. First, based on the six degrees of freedom model of the airship, he designed an adaptive backstepping sliding mode controller, used backstepping technology to obtain the desired speed, and used the adaptive sliding mode method to deal with unknown model uncertainty [2]. He We mainly considered the control design of flexible robotic arms. He moved the robotic arm to a specific desired angle, suppressed vibrations near the desired angle, and addressed the backlash nonlinearity in the actual system [3]. Combining DL and RPA technology can achieve a more intelligent and adaptive robot control system. Ensure the design of mechanical system is reasonable, including proper structure, strength and material selection, and adopt stable control algorithm to ensure the smooth operation of mechanical system.

The combination of DL and RPA can significantly improve the intelligence level and adaptive ability of robots, further expanding the field and scope of robot applications. Robotics technology further provides autonomy to perform critical tasks with minimal supervision in situations of limited resources. Chen Alvin I. has proposed a portable robot device that can introduce needles and catheters into deformable tissues such as blood vessels to autonomously draw blood or transport fluids. Robot intubation is driven by the prediction of a series of deep CNNS, which encode spatiotemporal information in multimodal image sequences to guide real-time servo [4]. Creating complex and autonomous soft robots requires these systems to have a reliable online ontology feel for 3D (Three dimensional) configurations through integrated soft sensors. Truby Ryan L. proposed a framework for predicting the 3D configuration of soft robots through DL using feedback from soft proprioceptive sensor skins [5]. However, they did not design a robot control system and conduct simulation experiments. The previously studied robot control system lacks the combination of deep learning and RPA, and lacks a controllable simulation environment for experimental analysis. Combining deep learning with robot process automation can learn from a large number of data, improve the robot’s perception and decision-making ability, and make it more suitable for different scenarios.

In order to improve the intelligence and adaptability of the robot control system, this article uses CNN and RPA technology to design the robot control system. By using CNN, robots can more accurately identify and understand their surroundings, including objects, obstacles, people and other key features. RPA technology allows robots to perform tasks automatically without human intervention. This includes identifying task requirements, planning and executing tasks, monitoring processes and automatically correcting errors. This improves the autonomy of the robot, especially when performing repetitive tasks. Based on the experimental results, it can be concluded that the performance of the robot control system based on CNN and RPA technology is very good, and the safety and reliability are also not very high. The innovation of this article is that DL algorithms can extract useful features and information from massive data, thereby improving data utilization and efficiency. At the same time, they can automatically process data and reduce manual intervention. Convolutional neural network and robot process automation technology are used to optimize the robot control system. CNN is used for image recognition to improve the robot’s perception ability, while RPA is used for task automation. The experiment evaluates the performance of the system, including the accuracy of perception and decision-making, time efficiency, motion trajectory, security and reliability. The innovation of this method lies in the combination of deep learning and automation technology, which provides a more intelligent, efficient and reliable control system for the application of robots in various fields.

2 DL and RPA technology

2.1 DL algorithms

CNN is a commonly used DL algorithm, commonly used in fields such as image recognition and object detection. CNNS are also widely used in robot control systems, which can improve the performance and accuracy of robot control systems by optimizing model training [6, 7].

Optimizing the model structure of CNN can improve the accuracy of robot control systems [8, 9]. When training CNN models, the complexity and accuracy of the model can be improved by increasing network depth, increasing the number and size of convolutional kernels, and other methods. In robot control systems, these optimizations can enable robots to more accurately identify and judge the surrounding environment and task requirements, thereby improving the accuracy and efficiency of robot task execution [10, 11].

Optimizing CNN training data can improve the performance of robot control systems [12, 13]. In robot control systems, training data often comes from sensors of the robot, such as cameras, LiDAR, etc. By increasing the quantity and quality of training data, the recognition and judgment ability of the robot can be improved, thereby improving the performance of the robot control system. At the same time, the training dataset can also be expanded through data augmentation and other methods to improve the generalization ability of the model.

Introducing reinforcement learning algorithms can optimize the training process of CNN models [14]. Reinforcement learning can guide the learning and optimization of models by rewarding and punishing robots for executing tasks, making robot control systems more intelligent and adaptive. For example, in robot navigation control, reinforcement learning algorithms can be used to optimize the training of CNN models, enabling the robot to complete navigation tasks more accurately.

Model compression and acceleration are also important methods for optimizing CNN model training [15]. In robot control systems, the size and computational complexity of models are often bottlenecks that limit the performance of robots. Therefore, model compression and acceleration techniques can be used to reduce the size and computational complexity of the model and improve the performance and efficiency of the robot control system. For example, models can be compressed through methods such as pruning, quantification, and distillation, and accelerated through techniques such as acceleration and model quantification [16, 17].

By optimizing the structure of the CNN model, training data, introducing reinforcement learning algorithms, and compressing and accelerating the model, the performance and accuracy of the robot control system can be improved, enabling the robot to perform tasks more intelligently and efficiently.

Assuming that the z-th to z + 3rd layers of a CNN are pooling layer, convolutional layer, and fully connected layer, respectively, the convolutional layers from the z-th pooling layer of the network to the z + 1st layer of the network are:

$a_{j}^{z + 1} = \sum_{i = 1}^{m_{j}} conv 2 (a_{i}^{z}, r_{ji}^{z + 1}, full) + y_{j}^{z + 1}, j = 1, . . ., m_{z + 1}$ (1)

$x_{j}^{z + 1} = g (l_{j}^{z + 1}) = g (a_{j}^{z + 1})$ (2) In the formula, $x_{j}^{z + 1}$ , $l_{j}^{z + 1}$ , $a_{j}^{z + 1}$ are the input, weighted input and activation values of the jth feature layer of the z + 1 layer of the network, g () is the activation function, conv2 is the convolution operation, that is, the operation of multiplying the sum of the corresponding elements, and full is the product operation that does not change the length and width of the feature layer. The pooling layers from the convolutional layer of the network layer z + 1 to the pooling layer of the network layer z + 2 include: $a_{j}^{z + 2} = downSample (x_{j}^{z + 1}), j = 1, . . ., m_{z + 2}$ (3) $x_{j}^{z + 2} = l_{j}^{z + 2} = a_{j}^{z + 2}$ (4) Among them, downSample () is a downsampling operation, which includes: $a_{j}^{z + 3} = \sum_{i = 1}^{m_{z + 2}} x_{j}^{z + 2} {* s}_{ji}^{z + 3} + y_{j}^{z + 3}, j = 1, . . ., m_{z + 3}$ (5) $x_{j}^{z + 3} = g (l_{j}^{z + 3}) = g (a_{j}^{z + 3})$ (6)

Among them, m_z+2 represents the number of neurons contained in a column expanded from layer z + 2, while $s_{ji}^{z + 3}$ and $y_{j}^{z + 3}$ represent the connection weights and biases between the i-th neuron in layer z + 3 and the i-th neuron in layer z + 2.

The above is the training model of a CNN from input layer A to output layer B. If (S, Y) is used to represent the parameters of the entire CNN, the entire CNN can be represented as: $b = g_{S, Y} (a)$ (7)

2.2 RPA technology

RPA technology is an automation technology that simulates human operations through robot programs, automatically executing repetitive, time-consuming, and error-prone tasks. In robot control systems, RPA technology is widely used, which can improve the efficiency and accuracy of robots, reduce operating costs, optimize business processes, and so on.

RPA technology can automate the execution of tasks by robots, thereby improving their work efficiency. For example, in industrial production, robots can automatically complete assembly, welding, spraying, and other tasks through RPA technology, improving the efficiency and output of the production line. In service robots, robots can automatically complete tasks such as transportation, cleaning, and meal delivery through RPA technology, improving service efficiency and quality [18].

RPA technology can reduce labor costs. In traditional production and service industries, a large amount of manpower is required to complete repetitive, time-consuming, and error-prone tasks. The use of RPA technology can replace these tasks, reduce labor costs, and improve the competitiveness and efficiency of enterprises [19, 20].

RPA technology can optimize business processes. In robot control systems, RPA technology can execute tasks in a programmatic manner, follow standardized processes, reduce human interference, reduce error rates, and improve work quality and efficiency. At the same time, RPA technology can also provide functions such as data analysis and decision support, helping enterprises better manage business processes and resources.

RPA technology can improve the intelligence level of robot control systems. By combining RPA technology with other artificial intelligence technologies such as DL, robot control systems can be made more intelligent and adaptive. For example, in robot navigation control, RPA technology can be used to automate tasks such as obstacle avoidance and path planning, and DL algorithms can help robots more accurately recognize and understand the surrounding environment and task requirements.

RPA technology has broad application prospects in robot control systems, which can improve the efficiency of robots, reduce costs, optimize business processes, and improve intelligence levels.

2.3 Components of robot control system

A robot control system is a complex system composed of multiple components, mainly including robots, sensors, actuators, controllers, communication protocols, and software.;Robots: Robots are the core part of the robot control system, completing tasks by performing preset tasks. Robots can be divided into various types, including industrial robots, service robots, medical robots, and so on.;Sensors: Sensors are an important component of the robot control system, used to collect information about the robot’s surrounding environment. According to different purposes, sensors can be divided into position sensors, force sensors, visual sensors, sound sensors, etc.Executor: Executor is another important component in the robot control system, used to control the motion of the robot. According to different purposes, actuators can be divided into electric motors, hydraulic motors, pneumatic motors, etc.

Controller: The controller is the core component of the robot control system, used to control the motion and operation of the robot.

Communication protocol: Communication protocol is an indispensable part of the robot control system, used to achieve data transmission and collaborative work between the robot and external devices.

Software: Software is the final component of the robot control system, used to achieve functions such as robot control, programming, and monitoring.

Improving the robot controller is an important task, aiming at improving the performance, intelligence and adaptability of the robot. The deep learning algorithm is used to improve the robot’s perception, decision-making and motion control capabilities. This will enable the robot to better understand the environment, identify objects, plan paths and perform tasks.

The construction of a robot control system requires selecting appropriate robots, sensors, actuators, controllers, communication protocols, and software based on specific application requirements for system design, development, testing, and debugging. At the same time, it is necessary to carry out system integration, optimization, and maintenance work to ensure the stability and reliability of the robot control system. The system architecture diagram is shown in Fig. 1.

Fig. 1

Interface functions and execution of robot control system.

3 Experimental design and robot control system based on DL and RPA

3.1 Experimental design

(1) Research background

In recent years, with the continuous development of robot technology and the continuous expansion of application scenarios, the research of robot control systems has become increasingly important.DL and RPA are two popular technologies that are widely used in the perception and decision-making of robot control systems, as well as in the execution of automated processes. Therefore, the research on robot control systems based on DL and RPA has important theoretical and practical value.

(2) Research purpose

This article aims to design a robot control system based on DL and RPA and explore its effectiveness and advantages in practical applications. Specifically, it includes the following aspects:

Deep learning methods can be used to perceive and recognize robots, improving their autonomous decision-making ability.

RPA technology can be used to automate robot execution processes, improving production efficiency and quality.

The integration of robot control systems with artificial intelligence, big data, and other technologies can be studied to explore the future development direction of robot control systems.

(3) Research content

System architecture design: This article would design a robot control system based on DL and RPA technology. The system architecture is divided into three layers: perception layer, decision-making layer, and execution layer. The perception layer is responsible for sensing and recognizing the surrounding environment of the robot; the decision-making layer processes and analyzes the perception results based on DL technology to make reasonable decisions; the execution layer is based on RPA technology to achieve the robot automation execution process.

(4) Algorithm implementation

The perception layer uses CNN algorithm for image recognition of the robot’s surrounding environment. The decision-making layer uses DL algorithms to process and analyze the perception results, making reasonable decisions. The execution layer adopts RPA technology to achieve the robot automation execution process.

(5) Experimental process

Use Python to import TensorFlow framework, set batch size to 32 and iteration times to 100, use cross entropy loss function as the loss function of Softmax classifier, and set the learning rate of the model to 0.0001. A large number of robot perception images are collected as data sets.

This article would design an experimental platform, including robots, cameras, sensors, and control systems. The experiment would be divided into two stages:

Perception and decision-making experiments: The robot would move and perceive independently in the laboratory, observe the surrounding environment, and make corresponding decisions, and test the robot’s perception and decision-making ability.

Perform experiments: The robot would complete a series of automated tasks in the laboratory to test the safety, stability, and reliability of the system.

(6) Research significance

The significance of this article is to explore the design and implementation methods of robot control systems based on DL and RPA technology, providing new ideas and methods for the research of robot control systems. It enhances the autonomous decision-making and automated execution capabilities of robots, providing technical support for achieving industrial automation and intelligent manufacturing. It promotes the integration of robot control systems with technologies such as artificial intelligence and big data, promoting the further development and application of robot control systems.

3.2 Experimental data

(1) Perception and decision experiment

The perception and decision-making experiment is an important part of this article, aiming to test the perception and decision-making abilities of robots. In the experiment, robots need to perceive and recognize the surrounding environment and make corresponding decisions based on the perception results.

The process of perception and decision-making experiments is as follows:

The robot began to move autonomously, while cameras and sensors began to work, collecting images and data of the surrounding environment.

The robot would transmit the collected images and data to the control system, which would use CNN algorithm for image recognition and data processing.

Based on reinforcement learning algorithm, the control system would make reasonable decisions based on the recognition and processing results, guiding the robot’s next action.

According to the instructions of the control system, the robot performs corresponding movements and actions, such as avoiding obstacles and searching for target objects.

The control system would monitor and evaluate the robot’s actions and adjust and optimize the algorithm model.

In order to test the perception and decision-making ability of robots, it is necessary to set some relevant parameters, such as image resolution, sensor sensitivity, and parameters of algorithm model. The specific parameter settings are as follows:

Image resolution: In order to improve the accuracy of image recognition, high-resolution cameras such as 1080P (Progressive) or higher should be used.

Sensor sensitivity: The sensitivity of the sensor should be adjusted according to the complexity of the environment. For example, in complex environments, the sensitivity of the sensor should be appropriately increased.

Parameters of the algorithm model: In order to improve the accuracy and efficiency of the algorithm model, it is necessary to adjust and optimize the parameters based on the experimental results, such as learning rate, discount factor, number of layers of the strategy network, and number of neurons.

The results of the perception and decision-making experiments would be evaluated and analyzed through the following indicators:

Image recognition accuracy: By comparing the recognition results with the actual situation, the image recognition accuracy can be calculated to evaluate the robot’s perception ability. The accuracy of image recognition refers to the measurement of an image recognition system’s ability to correctly identify objects or features in an image, which is the ratio of the number of correctly classified images to the total number of images.

Decision accuracy: By comparing the robot’s actions with the theoretically optimal actions, the decision accuracy can be calculated to evaluate the robot’s decision-making ability.

Time efficiency: By recording the time the robot completes tasks, the execution efficiency of the robot can be evaluated.

This article conducts five tests on the perception and decision-making abilities of robot control systems based on DL and RPA, and records the image recognition accuracy, decision-making accuracy, and time efficiency of the five experiments. In the five test experiments, the surrounding environment of the robot is different. The results and statistical characteristics are shown in Fig. 2, Table 1.

Fig. 2

Perception and decision experiment of robot control system based on DL and RPA.

Table 1

Statistical characteristics of five perception and decision-making experiments

Average value	Maximum value	Minimum value
Image recognition accuracy (%)	94.62	98.1	91.2
Decision accuracy (%)	87.5	91.4	83.5
Time efficiency(s)	176	192	160

From Fig. 2, it can be seen that the average image recognition accuracy of the perception and decision-making ability experiment of the robot control system based on DL and RPA is 94.62%. The highest is 98.1%, and the lowest is 91.2%. It indicates that the image recognition algorithm used in this experiment has high accuracy and can effectively identify target objects in the surrounding environment; the average accuracy of decision-making is 87.5%, with the highest being 91.4% and the lowest being 83.5%. The deep reinforcement learning algorithm used in this experiment can accurately make reasonable decisions based on environmental information; the average time efficiency is 176 seconds, with the longest being 192 seconds and the shortest being 160 seconds, indicating that the robot’s execution efficiency in this experiment is high and can complete tasks in a relatively short time. In the experiment on the perception and decision-making abilities of robot control systems based on DL and RPA, the robot has excellent perception and decision-making abilities, which can quickly and accurately perceive and recognize the surrounding environment, and make reasonable decisions to guide the robot’s actions. In practical applications, further optimization of algorithms and hardware devices is still needed to improve the flexibility and autonomy of robots. The average accuracy of image recognition is 94.62%, the average accuracy of decision-making is 87.5%, and the average time efficiency is 176 seconds. The robot control system based on DL and RPA can accurately perceive the environment and make decisions in a short time.

Experimental difficulties and solutions: The difficulty of perception and decision-making experiments lies in how to achieve autonomous perception and decision-making of robots. To solve this problem, advanced DL and reinforcement learning algorithms need to be adopted, and the algorithm model needs to be adjusted and optimized. At the same time, it is necessary to collaborate and optimize the hardware and software of the robot to improve its flexibility and autonomy. Perception and decision-making experiments are an important part of this article, which can test the perception and decision-making abilities of robots and lay the foundation for subsequent execution experiments. In the experiment, advanced algorithms and technologies need to be adopted, and the experimental parameters need to be reasonably set and adjusted to obtain accurate and reliable experimental results.

(2) Performing experiments

The execution experimental design for the research of robot control system based on DL and RPA is as follows:

Experimental purpose: It can evaluate the stability, safety, and system reliability of robot control systems based on DL and RPA.

Experimental equipment: It includes a robot, a computer, a robot control software, a laser sensor, a camera, and a set of obstacles.

Stability experiment: This article tests the motion trajectory deviation and oscillation amplitude of the robot, as shown in Fig. 3. During the experiment, the robot was placed on a flat ground and allowed to move along a predetermined path. The deviation between the actual motion trajectory of the robot and the predetermined path and the amplitude of the robot’s oscillation were recorded, as shown in Fig. 4. The standards for the deviation between the actual motion trajectory of the robot and the preset path, as well as the amplitude of human vibration, are 8 cm and 10°.

Fig. 3

Deviation of robot motion trajectory.

Fig. 4

Stability study in execution experiment.

From Fig. 4, it can be seen that the motion trajectory deviation and oscillation amplitude of the robot are maintained within a controllable range for one person, with the motion trajectory deviation not exceeding 5 cm and oscillation amplitude not exceeding 6°, belonging to the controllable range. This article shows that the stability of the robot control system based on DL and RPA is good by calculating the trajectory deviation and oscillation amplitude of the robot.

A safety experiment involves placing a robot in an environment with obstacles and personnel. This article uses laser sensors and cameras to monitor the environment and personnel around the robot. This article uses robot control software to move the robot along a predetermined path. This article records the distance between the robot and surrounding obstacles or personnel, as well as the movement speed of the robot, as shown in Fig. 5. The ratio standards for the distance between the robot and surrounding obstacles or people and the robot’s movement speed are known to be 30 cm and 1 m/s.

Fig. 5

Safety study during the execution of the experiment.

As shown in Fig. 5, the distance between the robot and surrounding obstacles or personnel, as well as the motion speed of the robot, are maintained within a safe range. The distance between the robot and surrounding obstacles or personnel does not exceed 20 cm, and the motion speed of the robot does not exceed 0.6 m/s. By calculating the distance and movement speed between the robot and its surrounding environment and personnel, the safety of the robot can be evaluated. The fact has proven that the robot control system based on DL and RPA has good security.

The system reliability experiment is to test the stability and reliability of the robot control system. During the experiment, the robot was asked to perform multiple movements according to the preset task, and then the operating time and number of failures of the robot control system were recorded. The experimental results are shown in Fig. 6. The standards for the operating time and failure frequency of the robot control system are known to be 30 hours and 0.3 times per hour.

Fig. 6

Reliability study in conducting experiments.

From Fig. 6, it can be seen that the running time and fault frequency of the robot control system based on DL and RPA are maintained within a reliable range, with running time not exceeding 25 hours and fault frequency not exceeding 0.2 times per hour. By calculating the running time and the number of failures of the robot control system, the reliability of the robot control system can be evaluated. Therefore, it can be concluded that the reliability of the robot control system based on DL and RPA is better.

In summary, through the experimental analysis of stability, safety, and system reliability, it can be concluded that the robot control system based on DL and RPA has good stability, safety, and system reliability.

4 Conclusions

The robot control system based on deep learning and RPA is a new type of robot control system, which uses deep learning algorithms and RPA technology to intelligently control and automate the execution of robots. This article aims to evaluate the stability, safety, and reliability of this system. After the experimental evaluation, the following conclusion was drawn: in the stability experiment, it was found that when the robot moves along the preset path, its deviation is small and the oscillation amplitude is relatively small. This indicates that the robot control system has high stability and can meet the requirements of robot motion. In the safety experiment, this article found that the robot can monitor the surrounding environment and personnel through laser sensors and cameras, and can avoid obstacles and personnel in a timely manner, ensuring the safety of the robot. Therefore, the robot control system has high safety. In the system reliability experiment, it was found that the robot control system runs longer and has fewer faults. This indicates that the robot control system has high reliability and can meet the requirements of long-term operation of the robot. The robot control system based on deep learning and RPA has shown a high level of stability, safety, and reliability. In the future, this control system is expected to be widely used in various robot applications, providing strong support for the intelligent development of robots. Experiments are carried out in a limited environment, which may not represent the performance of the robot in a complex and diverse real-world environment. Future research can consider more diverse scenarios and environmental conditions. In order to further improve the performance and efficiency of robots, further research and optimization of robot control algorithms and systems are needed.

References

Abbink

, David , Tom Carlson , Mark Mulder , Joost

C.F

, de Winter , Farzad Aminravan , Tricia

, Gibo , et al., A topology of shared control systems—finding common ground in diversity, IEEE Transactions on Human-Machine Systems 48.5 (2018), 509–525.

Wang Yueying , Weixiang Zhou , Jun Luo , Huaicheng Yan, , Huayan Pu , Yan Peng , Reliable intelligent path following control for a robotic airship against sensor faults, IEEE/ASME Transactions on Mechatronics 24.6(2019), 2572–2582.

He Wei , Xiuyu He , Mingfo Zou , Hongyi Li , PDE model-based boundary control design for a flexible robotic manipulator with input backlash, IEEE Transactions on Control Systems Technology 27.2 (2018), 790–797.

Chen Alvin

, Max

, Balter , Timothy

, Maguire ,, Martin

Yarmush ,, Deep learning robotic guidance for autonomous vascular access, Nature Machine Intelligence 2.2 (2020), 104–115.

Truby Ryan

, Cosimo Della Santina and Daniela Rus, Distributed proprioception of 3D configuration in soft, sensorized robots via deep learning, IEEE Robotics and Automation Letters 5.2 (2020), 3299–3306.

Chatzilygeroudis Konstantinos , Vassilis Vassiliades , Freek Stulp , Sylvain Calinon and Jean-Baptiste Mouret

R.K

, A survey on policy search algorithms for learning robot controllers in a handful of trials, IEEE Transactions on Robotics 36.2 (2019), 328–347.

Mandava

R.K

and Vundavilli

P.R

, An adaptive PID control algorithm for the two-legged robot walking on a slope, Neural Comput & Applic 32 (2020), 3407–3421.

Xiao Bing , Shen Yin , Exponential tracking control of robotic manipulators with uncertain dynamics and kinematics, IEEE Transactions on Industrial Informatics 15.2 (2018), 689–698.

Lindemann Lars , Dimos

, Dimarogonas , Control barrier functions for signal temporal logic tasks, IEEE Control Systems Letters 3.1 (2018), 96–101.

10.

Bilberg Arne , Ali Ahmad Malik , Digital twin driven human–robot collaborative assembly, CIRP Annals 68.1 (2019), 499–509.

11.

Zhang Binbin , Jun Wu , Liping Wang , Zhenyang Yu , Peng Fu , A method to realize accurate dynamic feedforward control of a spray-painting robot for airplane wings, IEEE/ASME Transactions on Mechatronics 23.3 (2018), 1182–1192.

12.

Joyee Erina

, Yayue Pan A fully three-dimensional printed inchworm-inspired soft robot with magnetic actuation, Soft Robotics 6.3 (2019), 333–345.

13.

Wang

, Mu

and Jin

, Control method of robot detour obstacle based on EEG, Neural Comput & Applic 34 (2022), 6745–6752.

14.

Zhang Shuang , Yiting Dong , Yuncheng Ouyang , Zhao Yin , Kaixiang Peng , Adaptive neural control for robotic manipulators with output constraints and uncertainties, IEEE Transactions on Neural Networks and Learning Systems 29.11 (2018), 5554–5564.

15.

Ruckstuhl

, Janji

and Rasp

, Training a convolutional neural network to conserve mass in data assimilation, Nonlinear Processes in Geophysics 28.1 (2021), 111–119.

16.

Lipps

, Herbst

, Klingel

, et al., Connectivity in the era of the (I)IoT: About security, features and limiting factors of reconfigurable intelligent surfaces, Discov Internet Things 3 (2023), 16.

17.

Abad-Segura Emilio, , et al., Abad-Segura Emilio, et al., Blockchain technology for secure accounting management: Research trends analysis, Mathematics 9.14 (2021), 1631.

18.

, Li

, Yang

and Liu

, A fast robotic arm gravity compensation updating approach for industrial application using sparse selection and reconstruction, Robotics and Autonomous Systems 149 (2022), 103971.

19.

Owusu Ebenezer , et al., A deep learning approach for loan default prediction using imbalanced dataset, IJIIT 19 (2023), 1–16.

20.

Qiao Liang , Yujie Li , Dongliang Chen , Seiichi Serikawa , Mohsen Guizani , Zhihan

Lv*

, A survey on 5G/6G, AI, and Robotics, Computers –Electrical Engineering 95 (2021), 107372.