Hindawi Template version: Apr19

Development of control algorithms for mobile robotics focused on

their potential use for FPGA-based robots

Andrés-David Suárez-Gómez,1 ANDRES A. HERNANDEZ ORTEGA,1,2

1 UNIVERSIDAD NACIONAL ABIERTA Y A DISTANCIA – Grupo de Investigación Byte

in Design
2
UNIVERSIDAD PEDAGOGICA Y TECNOLOGICA DE COLOMBIA.

Abstract

This paper investigates the development and optimization of control algorithms for mobile
robotics, with a keen focus on their implementation in Field-Programmable Gate Arrays
(FPGAs). It delves into both classical control approaches such as PID and modern techniques
including deep learning, addressing their application in sectors ranging from industrial
automation to medical care. The study highlights the practical challenges and advancements in
embedding these algorithms into FPGAs, which offer significant benefits for mobile robotics
due to their high-speed processing and parallel computation capabilities. Through an analysis
of various control strategies, the paper showcases the improvements in robot performance,
particularly in navigation and obstacle avoidance. It emphasizes the critical role of FPGAs in
enhancing the efficiency and adaptability of control algorithms in dynamic environments.
Additionally, the research discusses the difficulties in benchmarking and evaluating the
performance of these algorithms in real-world applications, suggesting a need for standardized
evaluation criteria. The contribution of this work lies in its comprehensive examination of
control algorithms' potential in FPGA-based mobile robotics, offering insights into future
research directions for improving robotic autonomy and operational efficiency.

Introducción
The design and optimization of control algorithms is often investigated to improve different
aspects of a robot's performance, such as path-following control for moving from one point to
another. There is a large amount of research on robot control algorithms and new approaches
are constantly being proposed. However, the problem arises of the difficulty in comparing the
results of these published research and evaluating the quality of the studies. Furthermore, in
robotics publications, performance evaluation criteria are often overlooked, making it difficult
to make an objective comparison of algorithms. Tests, whether in simulation or experimental,
are often limited to measuring the length of the path traversed or the time it takes the robot to
complete a task. In addition, there are few standard methods to evaluate the capabilities and
limitations of these systems in a comparable manner (Norton et al, 2019).

Research in this field is typically conducted in controlled laboratory settings to validate proofs
of concept and establish useful comparisons. However, it is important to keep in mind that the
results obtained may differ to some extent from the actual operation of a robot, as the latter is
characterized by the presence of uncertainty (Martins et al., 2020). In addition, there is a lack
of consensus on performance evaluation criteria, which often vary from one study to another.
This lack of consensus makes it difficult to compare the capabilities of navigation algorithms
and detracts from the rigor of the evaluation of advances in this field. Consequently, a
comprehensive evaluation system is lacking (Ren et al., 2020).

1
 Hindawi Template version: Apr19

Backgroung

Differential drive mobile robots: Differential drive mobile robots are a common category in
mobile robotics. These robots are characterized by having two independently driven wheels,
giving them great maneuverability and mechanical simplicity. Their ability to change the speed
and direction of each wheel independently allows for precise movements and turns on the spot.
A seminal text that addresses these types of robots is "Introduction to Autonomous Robots" by
Correll and Teller (2011). Previous research has explored specific control algorithms for these
types of robots, highlighting the importance of differential kinematics in path planning.

FPGA technology in robotics: FPGA technology plays a crucial role in robotics, especially
in the efficient implementation of control algorithms. FPGAs provide parallel processing and
reconfiguration capabilities, which is particularly beneficial in dynamic robotic environments.
To further understand the application of FPGAs in robotics, "Digital Design and Computer
Architecture" by Harris and Harris (2010) provides a detailed overview of FPGA architecture
and application. Previous research has demonstrated the successful use of FPGAs in the
implementation of embedded systems for real-time control of robots.

Control algorithms in robotics: Control algorithms are essential for the effective operation
of robots in diverse environments. "Principles of Robot Motion: Theory, Algorithms, and
Implementations" by Choset et al. (2005) is a comprehensive reference that addresses various
control algorithms used in mobile robotics. From path planning algorithms to feedback control
techniques, the existing literature highlights the need for adaptive and efficient algorithms.
Previous research has explored the application of techniques such as proportional-integral-
derivative (PID) control in stabilization and tracking of mobile robots.

Previous research and contributions: Review of previous research reveals that the field of
mobile robotics has experienced significant advances in control algorithms, application of
FPGA technologies, and differential drive robot design. For example, "Robotics Research: The
13th International Symposium ISRR" (2017) presents several contributions in the field of
autonomous navigation and control of mobile robots. Previous contributions have
demonstrated the effectiveness of heuristic algorithm-based path planning and implementation
of real-time control systems using FPGA in challenging robotic environments.

Hardware Architecture

Role of the FPGA in the robot:

FPGA plays a crucial role in the design and implementation of robotic systems, providing
significant flexibility and efficiency. The reconfigurability of FPGAs allows them to adapt to
changes in the operating environment and control requirements of the robot. This feature is
especially valuable in applications where the complexity of the control algorithm can vary
dynamically.

In the specific context of a robot, the FPGA can be used to implement real-time control
algorithms, improving the responsiveness and accuracy of the system. In addition, the parallel
architecture of FPGAs is ideal for executing multiple tasks simultaneously, which can be
essential in robotic applications that require fast sensory data processing and real-time decision
making.

2
 Hindawi Template version: Apr19

A highlight of the FPGA's role in robots is its ability to improve energy efficiency. The
implementation of control algorithms in configurable hardware allows the computational load
on the robot's main CPU to be reduced, which can translate into lower power consumption and
increased autonomy.

Previous research has successfully explored the integration of FPGAs into task-specific robots,
demonstrating remarkable improvements in processing speed and system efficiency.
Specialized literature, such as "FPGA-Based Implementation of Signal Processing Systems"
by Woods et al. (2008), provides a deeper understanding of how FPGA can contribute to the
overall performance of a robot.

FPGA Implementation
Basic Concepts:

FPGA (Field-Programmable Gate Array): An FPGA is a digitally configurable

semiconductor device that allows designers to program or configure the digital logic and
interconnect within the chip according to their specific needs.

Programmable Logic Array (PL): The programmable logic array is the core of an FPGA
and consists of an arrangement of logic blocks and connections that can be programmed to
perform specific tasks.

Logic Blocks: Logic blocks are basic units that contain logic gates and flip-flops. These
blocks are configured to implement desired logic functions.

Interconnect: The interconnect consists of the paths that connect the logic blocks and other
elements within the FPGA, enabling communication and data transfer.

Configuration: Programming an FPGA involves establishing the functionality of its logic

blocks and interconnect using a configuration file. This process determines how the FPGA
will perform specific tasks.

FPGA Advantages:

Flexibility and Reconfigurability: The main advantage of FPGAs is their ability to be

reconfigured according to the designer's needs. This allows the hardware to be adapted to
different applications without the need to change the physical components.

High Performance: FPGAs can offer significant performance compared to other

programmable technologies, especially in applications requiring parallelism and real-time
data processing.

Rapid Development: Reconfigurability enables rapid prototyping and proof-of-concept

development. Designers can easily make iterations and adjustments without having to
redesign the hardware.

Custom Parallelism: The parallel architecture of FPGAs enables highly parallel algorithms
to be implemented, improving efficiency and processing speed compared to sequential
implementations.

3
 Hindawi Template version: Apr19

Energy Efficiency: By enabling the specific implementation of the hardware required for a
task, FPGAs can be more energy efficient compared to general-purpose solutions, reducing
power consumption in embedded systems and mobile devices.

Adaptability to Change: The ability to reconfigure makes it easier to adapt to changes in

design requirements, which is especially valuable in environments where requirements may
evolve over time.

Implementation of Specific Functions: FPGAs are ideal for implementing specific

functions efficiently, such as signal processing algorithms, image processing, and embedded
system control.

Materials and Methods

Normally controlling the inputs of a monocycle model is about applying the traditional
feedback PID controller and selecting the appropriate input u = (νω)𝑇 , given by eq:
1 𝑑𝑒 (𝑡)
𝑈 (𝑡) = 𝑃𝐼𝐷 (𝑒) = 𝐾𝑝 𝑒(𝑡) + 𝐾𝐼 ∫ 𝑒(𝜏) 𝑑(𝜏) + 𝐾𝐷 (1)
0 𝑑𝑡

In the context of each task detailed below, the term 'e' refers to the error between the desired
value and the value obtained as a result. The constants Kp, Kl and KD represent the
proportional, integrative, and derivative gains respectively, while 't' refers to time. The control
gains used in this study are determined by adjusting different values in order to achieve
satisfactory responses. If the vehicle travels at a constant speed, ν = ν0, then the control input
will only change along with the angular velocity, ω, in this manner:

𝑤 = 𝑃𝐼𝐷 (𝑒) (2)

Results and Discussion

Space-time Criteria

Performance criteria related to space-time dimensions are widely used and allow a quantitative
evaluation and comparison of results in real experiments or in simulation. In the article by
Munoz et al. (2014), several typical criteria in navigation and obstacle avoidance are described,
such as mission success, robustness in narrow spaces, path length, time required to complete
the mission, control periods, average distance to target, distance to obstacles, and smoothness
of trajectory, among others.

Among these criteria, those related to space-time dimensions are the simplest and most used.
An optimal trajectory from the point of view of reaching the goal is one that follows a straight
line of the minimum possible length and without curvature between the starting point (xi, yi)
and the arrival point (xn, yn), traversed in the shortest time. This approach assumes linearity
and constant velocity of the robot in its trajectory towards the goal (Munoz-Ceballos, N. D. et
al., 2022).

The Dynamic Window Approach (DWA) is a well-known collision avoidance navigation

algorithm. It was initially proposed by Dieter Fox and his team. DWA works in real time and
reacts to changing situations as they occur. In recent years, the DWA cost function has

4
 Hindawi Template version: Apr19

undergone several extensions and improvements. This approach determines safe and optimal
translation (v) and rotation (w) speeds directly by creating velocity profiles that take into
account robot dynamics and speed and acceleration constraints. The search for suitable
velocities mainly involves three subspaces, including the space of possible values of v
(Mohammadpour et al, 2021).
The space of possible velocities is divided into three subspaces according to the kinematic
constraints of the robot:

Space of Possible Velocities (Vs): This subspace considers the kinematic constraints of the
robot and represents all velocities that are physically possible for the robot based on its
mechanical characteristics.

Admissible Velocity Space (Va): This subspace considers the velocities that allow the robot to
stop without colliding with an obstacle. These velocities are restricted by the robot's ability to
stop safely and without collision.

Space of Possible Speeds with Consideration of Robot Acceleration Limitations (Vd): This
subspace considers the acceleration limitations of the robot. It represents the possible velocities
considering the robot's limited ability to change its velocity quickly due to its acceleration
constraints.
𝑉𝑟 = 𝑉𝑠 ∩ 𝑉𝑎 ∩ 𝑉𝑑 (3)

Where Vr is the optimal velocity search space, which is selected by maximizing the following
objective function:
𝐺 (𝑣, 𝑤) = ∝∗ ℎ(𝑣, 𝑤) + 𝛽 ∗ 𝑑(𝑣, 𝑤) + 𝛾 ∗ 𝑣𝐹 (𝑣, 𝑤) (4)

Where "h" measures the alignment of the robot with the target direction, "d" represents the
distance to the nearest obstacle, and "vF" is the forward velocity of the robot. The values of α,
β and γ are adjustable constants that determine the relative importance of these three measures
in the objective function. In summary, the DWA method generates numerous possible local
online trajectories and then selects the most suitable one based on the objective function.
Finally, the most suitable velocity is run to follow the selected local trajectory.

Success in reaching the goal

This criterion is generally given in terms of percentage (%), it consists of counting the
percentage of success in completing a navigation mission with respect to the total number of
attempts. Some researchers also set a time limit, but sufficient to successfully complete the
given navigation mission, this in order to rule out tests where the robot gets stuck navigating
in an endless loop (McGuire et al, 2019).An additional challenge for control algorithms is their
performance in navigating through narrow passages or corridors, therefore, an additional
criterion to consider may be the robustness in narrow spaces: number of narrow passages
successfully traversed.

As an example it is used in navigation robots that must find the exit to a maze, in such situation
the triangulation of the position is done by the configuration of the solved triangle, where we
want to calculate the height (h) of the triangle with sides 'a' and 'b' and angle β. 'c' is the side of
the triangle that will be unknown, so a formula will be developed that will only use 'a,' 'b' and
β.

5
 Hindawi Template version: Apr19

𝑐. ℎ
𝐴= (5)
2

𝑎. 𝑏. sin 𝛽
𝐴= (6)
2
The cosine theorem is applied to triangulate the height and develop the triangle by replacing
the above equations.

𝑎. 𝑏. sin 𝛽
ℎ= (7)
√𝑎2 + 𝑏 2 − 2. 𝑎𝑐𝑜𝑠𝛽

Figure 1. Diagram to explain the wall-tracking paradigm for a local address on the right-hand side with "(g)"
representing the corresponding state machine. "OR" stands for "out of range". Taken from (McGuire et al, 2019).

Time reaching the target

Execution time is a fundamental criterion used in most articles comparing algorithms. It

consists of measuring the time the robot takes to reach the goal. In simulations of deterministic
systems, the same result is obtained under the same simulation conditions. However, to bring
the simulation closer to reality, where there are factors such as battery wear, friction between
the wheels and the ground, environmental conditions, among others, noise can be introduced
into the system. This is achieved by adding disturbances in the sensor readings or in the control
signals to the actuators. In simulation, random number generators can be used to model the
error or noise.

Another related criterion is the number of processing cycles, which is equivalent to the
approximate number of operations performed to complete a mission. It is important to note that
different types of robots may have different types of processors, i.e., different computational
capabilities. This influences the total time the robot takes to complete the mission and the
comparison of algorithms using this criterion (Tai et al., 2020).

Control periods

Control periods refer to the number of times the planner makes decisions to reach the goal.
This measure is related to the number of iterations or steps the robot needs to complete the

6
 Hindawi Template version: Apr19

mission. If the robot moves at a constant linear velocity (v), the control periods provide an
estimate of the time taken to complete the mission. The greater the number of control periods,
the greater the number of decisions made and, potentially, the greater the time required to reach
the goal.

Error-based Criteria

In a control system, error is defined as the difference between the controlled variable (also
known as the process variable) and the reference value or set-point. In a control system, the
objective is that the error tends to zero, indicating good system performance.

One way to evaluate the performance of a control system is to quantify the cumulative error.
In the case of a mobile robot, the cumulative error provides a numerical measure of how "good"
the performance of a specific controller is for the robot's drive or steering system.

In discrete-time controllers, it is necessary to know the error e(nT) at each sampling instant,
where T is the sampling period and fs is the sampling frequency. The sampling frequency has
a direct impact on the accuracy of the measurements and the ability of the controller to detect
and correct the error over time.

Error-based or error-integral performance criteria have a well-established theoretical

foundation in control systems, and are widely used to evaluate and improve controller
performance (Domański, 2020). These criteria allow the system behavior to be analyzed in
terms of error reduction and control stability.

Final error

Final error refers to the discrepancy between the final position of the vehicle and the end point
of an established reference trajectory. It is calculated by measuring the distance between the
final position of the vehicle and the desired end point of the reference trajectory. This
measurement is especially useful in robotic underwater vehicles, as it allows detecting if the
vehicle has lost the trajectory halfway through the tracking or if it has reached an incorrect end
position (Perez et al., 2018).

As an example, the article mentions that by means of a developed platform, it is possible to

automatically evaluate the performance of the obtained solutions using a specific metric. In
this context, a typical metric in control algorithms for trajectory tracking is used: the integrated
squared error (ISE) and the final error. The ISE is calculated as the sum of the distances to the
ideal trajectory, which is two meters above the pipe, over time. This measure is inversely
related to tracking quality, as it considers both time-spent and tracking accuracy.

The rate of increase is significantly higher as the distance between the vehicle and the optimal
trajectory increases, but it also penalizes excessively slow trajectory tracking. In addition, it is
essential to determine whether the end of the pipeline has been reached. To achieve this, the
distance between the end position of the vehicle and the end of the pipeline is calculated. In
this way, it is possible to identify if the vehicle has lost the pipeline during the middle of the
tracking or if it has incorrectly detected the end. Taking these measurements into consideration,
the final evaluation is determined by equation 7.

7
 Hindawi Template version: Apr19

𝑚𝑒𝑎𝑛𝑒𝑟𝑟𝑜𝑟 2 𝑡𝑖𝑚𝑒 − 𝑟𝑒𝑓

𝑒𝑣𝑎𝑙𝑢𝑎𝑡𝑖𝑜𝑛 = (1 − 𝑒𝑟𝑟𝑜𝑟 2 ) ∗ (0,1 − + (7)
0,1 100

Where "error" represents the final error, "mean error" is the average error over the tracking
calculated from the ISE, "time" indicates the time required to perform the intervention, and
"ref" refers to the time reference for each scenario, determined as a function of distance
traveled, turns and changes in altitude. The first term of the equation evaluates the final position
of the vehicle, penalizing those situations in which the vehicle is far from the target. The second
term evaluates the average error along the trajectory. The last term is a bonus that favors fast
tracking and penalizes slow tracking based on the complexity of the pipeline path.
In general controller design, there are commonly used performance criteria, such as indices
involving the integral of the error (Suarin et al., 2019). These criteria are based on cumulative
error and can be applied to reference trajectory tracking, indicating the error along the entire
path between the reference trajectory and the actual trajectory followed by the robot. These
indices are also used in position control, distance, orientation, multiple robot formation, among
others (Caruntu et al., 2019) (Farias et al., 2020). The lower the error, the better the trajectory
traversed and, consequently, the better the control algorithm.

Safety Criteria

Related research addressing performance criteria in robot control and its safety in navigation
is described in the study conducted by Marvel and Bostelman in 2014. These performance
criteria focus on the safety of the robot while moving along a given trajectory, taking into
account factors such as the distance between the vehicle and the obstacles encountered in its
path, as well as the number of collisions occurring during navigation (Munoz et al., 2014).
These investigations seek to ensure safe travel and avoid possible accidents or damage during
the operation of the robot.

The average distance to obstacles during the entire navigation mission is another performance
criterion used in robot control. This criterion makes it possible to evaluate how close or far the
robot is from obstacles in its environment during the entire trajectory followed. In an
unobstructed environment, this maximum value will be higher, since the robot can move freely
without restrictions. On the other hand, if the average distance to obstacles index deviates less
from the maximum value, it means that the path followed by the robot passed through more
obstacle-free zones, indicating better performance in terms of avoidance and safe navigation.
This criterion is important to avoid collisions and ensure a clear and obstruction-free path for
the robot during its navigation mission.

The minimum average distance to obstacles is another performance criterion used in robot
control to evaluate safety during a navigation mission. It averages the minimum distance value
measured by each of the robot's n sensors. This criterion provides an idea of the risk taken
during the mission in terms of proximity to obstacles.

In unobstructed environments, where there are no obstacles close to the robot, it will be fulfilled
that the minimum average distance to obstacles will be equal for all sensors. This indicates that
the robot has maintained a safe distance and has not been exposed to significant risk of collision
or contact with obstacles.

In contrast, in environments with obstacles, the smaller the minimum mean distance to
obstacles, the greater the risk that has been taken during the mission, since the robot has been

8
 Hindawi Template version: Apr19

closer to the obstacles. Therefore, a lower value of this criterion indicates a higher probability
of collisions or contact with obstacles.

Energy Consumption

Energy consumption is a key aspect in mobile robot applications, as it influences the autonomy
of the robot, that is, the time it can function optimally before running out of energy. In recent
years, special attention has been paid to this topic (Stefek et al, 2020). If a robot does not meet
energy consumption requirements, such as the ability to operate independently and the ability
to recharge, its performance, operating time and autonomy are limited (Heikkinen et al, 2018).

Mobile robots rely heavily on batteries as a power source, but these have limited energy
capacity. As a result, the robot's operating time is often short, which may be insufficient for
tasks or missions that require more time and energy to complete. Increasing run time by using
more batteries or directing the robot to a charging station can increase the cost or size of the
system, which can cause control issues. An alternative is to improve the energy efficiency of
the robot, reducing its energy consumption (Armah et al, 2016).

References

Correll, N., & Teller, S. (2011). Introduction to Autonomous Robots.

Kelly, A. (2013). Mobile Robotics: Mathematics, Models, and Methods.

Siciliano, B., & Sciavicco, L. (2008). Robotics: Modelling, Planning and Control.

Harris, D., & Harris, S. (2010). Digital Design and Computer Architecture.

Woods, R., Pearson, C., & Miller, S. (2008). FPGA-Based Implementation of Signal
Processing Systems.

ISRR (2017). Robotics Research: The 13th International Symposium ISRR.

Hsu, D., Dissanayake, G., & Laumond, J. P. (Eds.). (2006). Field and Service Robotics:
Recent Advances in Research and Applications.

Armah S., Yi S., Lebdeh T. A.,2016.Implementation of autonomous navigation algorithms on two-wheeled

ground mobile robot.Amer. J. Eng. Appl. Sci. 78, 36-48. DOI:10.3844/ajeassp. 2014.149.164.

Caruntu C., Copot C., Lazar C.,Keyser R. D., 2019. Decentralized Predictive Formation Control for Mobile
Robots without Communication. In:IEEE 15th Int.Conf.on Control and Automation (ICCA),pp. 555-
560.Edinburgh, United Kingdom.DOI:10.1109/ICCA.2019.8899610.

Domański, P. D., 2020. Control Performance Assessment: Theoretical Analyses and Industrial Practice.
Springer, Warsaw.DOI: 10.1007/978-3-030-23593-2

Farias G., Garcia G., Montenegro G., Fabregas E., Dormido-Canto S., Dormido S.,2020.Reinforcement
Learning forPosition Control Problem of a Mobile Robot. IEEE Access8,152941-152951.
DOI:10.1109/ACCESS.2020.3018026.

Heikkinen J. E., Minav T. and Serykh E. V.,2018.Mobile robot qualification metrics. In:IEEE Conference of
Russian Young Researchers in Electrical and Electronic Engineering (EIConRus), pp. 878-881,Russia.
DOI:10.1109/EIConRus.2018.8317228.

9
 Hindawi Template version: Apr19

Martins, O., Adekunle, A., Adejuyigbe, S., Adeyemi, O., Arowolo, M.,2020. Wheeled Mobile Robot Path
Planning and Path Tracking Controller Algorithms: A Review. Journal of Engineering Science and
Technology Review 13(3),152 –164. DOI:10.25103/jestr.133.17

McGuire K.N., de Croon G.C.H.E., Tuyls K., 2019. A comparative study of bug algorithms for robot navigation,
Roboticsand Autonomous Systems 121, 103261. DOI:10.1016/j.robot.2019.103261

Mohammadpour M., Zeghmi L., Kelouwani S., Gaudreau M., Amamou A., Graba M.,2021. An
Investigation into the Energy-Efficient Motion of Autonomous Wheeled Mobile Robots. Energies 2021, 14,
3517. DOI: 10.3390/en14123517

Munoz, N.D., Valencia, J., Alvarez, A.,2014. Simulation and Assessment Educational Framework for
Mobile Robot Algorithms. J. Autom. Mob. Robotics Intell. Syst., 8, 53-59. DOI:10.14313/JAMRIS_1-2014/7

Norton A., Gavriel P., Yanco H., 2019. A Standard Test Method for Evaluating Navigation and Obstacle
Avoidance Capabilities of AGVs and AMRs.Smart and Sustainable Manufacturing Systems 3, (2).106-
126. DOI:10.1520/SSMS20190028

Pérez J., Fornas D, Marín R., Sanz P. 2018. UWSim, un Simulador Submarino Conectado a la Nube como
Herramienta Educacional. Revista Iberoamericana de Automática e Informática Industrial 15, 70-78.
DOI:10.4995/riai.2017.8827

Ren Y., Wang J., Zheng X., Zhao Q., Ma J., Zhao L.,2020. Research on multidimensional evaluation of
tracking control strategies for self-driving vehicles. Advances in Mechanical
Engineering.DOI:10.1177/1687814020912968

Stefek A., Pham T., Krivanek V.,Pham K., 2020. Energy Comparison of Controllers Used for a Differential
Drive Wheeled Mobile Robot.in IEEE Access, vol. 8, pp. 170915-170927 DOI:10.1109/ACCESS.2020.3023345.

Suarin N.A.S., Pebrianti D., Ann N.Q., Bayuaji L., Syafrullah M., Riyanto I.,2019.Performance Evaluation of
PID Controller Parameters Gain Optimization for Wheel Mobile Robot Based on Bat Algorithm and
Particle Swarm Optimization. Lecture Notes in Electrical Engineering, vol 538. DOI:10.1007/978-981-13-3708-
6_27

Tai J., Phang S., Wong Y., 2020. Optimized autonomous UAV design with obstacle avoidance capability.AIP
Conference Proceedings 2233, 020026.DOI:10.1063/5.0001372

10