ALGORITHMIC STUDY ON OPTIMAL PATH PLANNING FOR A SANITIZATION

ROBOT

1. Introduction

Overview of the Project Shift from Robot Development to Algorithmic Path Planning

The initial focus of this project was on creating a physical robot designed for hospital
sanitisation. However, the emphasis has now transitioned to developing sophisticated algorithms
that determine the robot's movement paths. This shift recognises that while the robot's hardware
is essential, the efficiency and effectiveness of its cleaning tasks heavily depend on the
intelligence of its path planning. By concentrating on algorithmic development, we aim to
enhance the robot's ability to navigate complex hospital environments, ensuring thorough and
efficient disinfection.

Importance of Optimal Path Planning in Hospital Sanitisation

In hospital settings, maintaining a sterile environment is crucial to prevent the spread of

infections. Autonomous robots equipped with disinfection capabilities, such as UV-C light, have
emerged as valuable tools in this endeavour. However, the mere presence of such robots is not
sufficient; their effectiveness is largely determined by how well they cover the areas requiring
disinfection. Optimal path planning ensures that the robot covers all necessary surfaces without
unnecessary repetition or omission, maximising disinfection efficiency and reducing operational
time. Efficient path planning also conserves energy and extends the robot's operational lifespan,
making it a cost-effective solution for healthcare facilities. Studies have shown that
implementing advanced path planning algorithms can significantly enhance the performance of
disinfection robots in complex environments.

Use of UV-C Bulbs and Constraints Associated with Them

UV-C light, with wavelengths between 200-280 nanometres, is effective in neutralising a wide
range of pathogens, including bacteria and viruses, by disrupting their DNA and RNA. This
makes UV-C bulbs a powerful tool for hospital disinfection. However, their use comes with
specific constraints. Firstly, UV-C light can be harmful to human skin and eyes, necessitating that
disinfection robots operate in unoccupied spaces or have safety measures to prevent human
exposure. Secondly, the intensity and duration of UV-C exposure must be carefully calibrated;
insufficient exposure may not effectively neutralise pathogens, while excessive exposure can
degrade materials and surfaces within the hospital environment. Additionally, UV-C light has
limited penetration abilities, meaning it cannot disinfect shaded areas or those not in the direct
line of sight. Therefore, the robot's path planning must account for these limitations, ensuring
that all critical surfaces receive adequate exposure.

Objectives of the Study

This study aims to develop an optimal path planning strategy for autonomous hospital
disinfection robots. The specific objectives include:

 Developing algorithms that enable complete coverage of designated areas, ensuring no

spots are missed during the disinfection process.
 Minimising the total path length and operational time, thereby enhancing energy
efficiency and reducing wear on the robot.

2. Introduction to Path Planning in Robotics

Path planning is a crucial aspect of robotics, involving the determination of a collision-free route
from a robot's starting point to its destination. This process ensures that robots can navigate
through environments, whether static or dynamic, while avoiding obstacles and efficiently
reaching their goals. Effective path planning is essential for various applications, including
autonomous vehicles, service robots, and industrial automation.

2.1. Key Challenges in Path Planning for Sanitisation

In the context of sanitisation, robots are deployed to disinfect areas, making path planning
particularly challenging due to several factors:

 Dynamic Environments: Sanitisation robots often operate in spaces with moving

obstacles, such as people or other machinery, requiring real-time path adjustments.

 Complete Coverage: Ensuring that the robot covers every surface area without omission
is vital for effective disinfection.

 Time Constraints: Sanitisation tasks may need to be completed within specific time
frames to minimise disruption, necessitating efficient path planning.

 Safety Considerations: Robots must operate without posing risks to humans, especially
when using potentially harmful disinfectants like UV-C light.

2.3 Existing Path Planning Algorithms:

Several algorithms are employed in robotic path planning, each with its advantages and
limitations:

 A Algorithm*
The A* (A-star) algorithm is a widely used pathfinding and graph traversal technique that
aims to find the shortest path between two points. It combines features of Dijkstra's
algorithm and a heuristic approach to efficiently navigate through a map or grid. A*
evaluates paths by combining the cost to reach a point and an estimated cost to the
destination, ensuring an optimal and efficient route. This algorithm is particularly
effective in static environments where the terrain is known and unchanging.
 Dijkstra’s Algorithm
Dijkstra's algorithm, developed by Edsger W. Dijkstra, is a classic method for finding the
shortest paths between nodes in a graph. It systematically explores all possible paths from
the starting point, ensuring that the shortest path to each node is found. While thorough,
Dijkstra's algorithm can be computationally intensive, especially in large graphs, as it
does not use heuristics to guide its search. It is most suitable for scenarios where all edge
weights are non-negative and a comprehensive pathfinding solution is required.
  Rapidly-exploring Random Trees (RRT)
Rapidly-exploring Random Trees (RRT) are designed to efficiently explore large, high-
dimensional spaces by randomly building a space-filling tree. This approach is
particularly useful in robotics for path planning in complex environments with obstacles.
RRT algorithms incrementally construct paths by randomly sampling points in the space
and connecting them to the nearest existing tree node, ensuring rapid exploration of
feasible paths. They are especially advantageous in scenarios where the environment is
dynamic or not fully known in advance.
 Genetic Algorithms (GA)
Genetic Algorithms are optimisation techniques inspired by the process of natural
selection. They work by evolving a population of candidate solutions over time, using
operators such as selection, crossover, and mutation. In path planning, GAs can be
employed to find optimal or near-optimal paths by iteratively improving a set of potential
solutions based on a defined fitness function. This method is particularly useful in
complex search spaces where traditional algorithms may struggle to find efficient
solutions.
 Simultaneous Localization and Mapping (SLAM)
Simultaneous Localization and Mapping (SLAM) is a process by which a robot or
autonomous system constructs a map of an unknown environment while simultaneously
keeping track of its own location within that map. This dual capability is crucial for
effective navigation and path planning in environments where prior maps are unavailable
or incomplete. SLAM combines data from various sensors (such as LiDAR, cameras, and
inertial measurement units) to build and update the map in real-time, allowing the system
to navigate accurately and avoid obstacles.

2.4 Application of These Algorithms in Robotics and Sanitisation

In robotics, path planning algorithms are essential for enabling autonomous navigation and task
execution. For instance, the A* algorithm has been applied to robot navigation in structured
areas, demonstrating robust performance in path optimisation and obstacle avoidance in static
environments. However, its lack of adaptability to dynamic settings limits its application in more
realistic and changing scenarios.

Rapidly-exploring Random Trees (RRT) have been utilised for path planning in complex
environments, such as space robotic manipulators, where obstacle avoidance is critical. An
improved RRT algorithm has been proposed to guide space robots, enhancing their ability to
navigate safely and efficiently in cluttered spaces.

In the context of sanitisation, particularly with UV-C disinfection robots, path planning
algorithms ensure comprehensive coverage of areas to be disinfected. A study introduced an
autonomous UV-C disinfection strategy based on a dynamic irradiation map, which indicates the
amount of energy applied in each region. This approach allows the robot to adapt its path and
speed in real-time, ensuring effective disinfection even in unknown environments.

2.5 Limitations of Traditional Path Planning for UV-C-Based Sanitisation

Traditional path planning algorithms, while effective in many applications, face challenges when
applied to UV-C-based sanitisation tasks. One significant limitation is their performance in
dynamic or unstructured environments. Algorithms like A* and Dijkstra's are designed for static
settings and may not adapt well to changes, such as moving obstacles or alterations in the
environment, which are common in real-world scenarios.

Moreover, these algorithms often do not account for the specific requirements of UV-C
disinfection, such as the need for adequate exposure time and coverage to ensure effective
sanitisation. A study highlighted that traditional path planning might not guarantee disinfection
of the entire area, potentially leaving surfaces insufficiently irradiated due to inadequate path
planning.

To address these challenges, researchers have developed specialised algorithms that consider the
unique aspects of UV-C disinfection. For example, an improved RRT* algorithm has been
proposed for disinfecting robot path planning, which enhances exploration speed and reduces the
average initial path search time, ensuring more efficient and thorough coverage.

3. Unique Constraints for UV-C Sanitisation Robots

Ultraviolet-C (UV-C) sanitisation robots are designed to disinfect environments by emitting UV-
C light, which is effective in inactivating various pathogens. However, their operation involves
unique challenges that influence their design and functionality.

3.1 UV-C Light Properties and Their Impact on Path Planning

UV-C light, with wavelengths between 200–280 nanometres, is highly effective at neutralising
microorganisms by damaging their DNA or RNA. However, UV-C radiation has limited
penetration abilities and can be obstructed by materials like glass or plastic. This necessitates
careful path planning to ensure that all surfaces receive adequate exposure. Robots must navigate
around obstacles and adjust their positions to minimise shadows and ensure comprehensive
disinfection. Advanced path planning algorithms are employed to map environments accurately,
allowing robots to determine optimal routes that maximise UV-C exposure while avoiding
obstacles. These algorithms consider factors such as room geometry, object placement, and
potential shadowed areas to create effective disinfection paths.

3.2 Distance and Intensity Requirements for Effective Sanitisation

The effectiveness of UV-C sanitisation is influenced by the distance between the UV-C source
and the target surface, as well as the intensity and duration of exposure. The intensity of UV-C
radiation decreases with distance, following the inverse square law. Therefore, surfaces closer to
the UV-C source receive higher doses, while those farther away receive less. To ensure effective
disinfection, robots must control their proximity to surfaces, adjusting their paths to maintain
optimal distances. Additionally, exposure time must be managed to deliver the necessary UV-C
dose for pathogen inactivation. This requires precise control of robot speed and positioning,
ensuring that each area receives sufficient exposure without unnecessary energy expenditure.
3.3 Field of View (FoV) Considerations

The Field of View (FoV) of a UV-C sanitisation robot refers to the area that the UV-C light can
effectively cover at any given time. A wider FoV allows the robot to disinfect larger areas with
each pass, but it also requires careful management to avoid shadowed regions where pathogens
might survive. Robots must be equipped with sensors and mapping technologies to identify
objects that could block UV-C light, adjusting their positions and angles to ensure
comprehensive coverage. Strategic planning of the robot's movements, considering its FoV, helps
in reducing disinfection time and ensuring all surfaces are adequately exposed.

3.4 Surface Geometry and Environmental Constraints

The geometry of surfaces and environmental factors significantly impact the effectiveness of
UV-C sanitisation. Complex shapes, textured surfaces, and tight spaces can create shadows or
areas where UV-C light exposure is insufficient. Robots must be capable of navigating diverse
environments, adjusting their orientation and distance to ensure all areas are reached. This may
involve using articulated arms or adjustable mounts to position the UV-C source optimally.
Environmental constraints, such as the presence of sensitive equipment or materials that could be
degraded by UV-C exposure, require robots to adapt their paths and possibly modulate UV-C
intensity to prevent damage. Advanced sensors and real-time mapping enable robots to detect
and respond to these challenges dynamically, ensuring effective and safe sanitisation.

3.5 Energy Efficiency Considerations

Balancing disinfection efficacy with energy consumption is vital. Efficient path planning can
minimise the distance travelled and the time spent, thereby reducing energy usage. This involves
calculating the most effective route that covers all necessary areas with the least amount of
movement and time. Additionally, adjusting the robot's speed based on the required exposure
time for different areas can help in conserving energy while ensuring effective disinfection.
Incorporating these considerations into UV-C sanitisation path planning enhances both the
effectiveness and efficiency of disinfection processes.

4. Mathematical Formulation

We will show the mathematical foundations essential for optimising robotic disinfection
processes. Our focus encompasses defining the optimisation problem, exploring geometric and
kinematic considerations, modelling illumination, and representing the environment through
graph theory.

4.1 Optimisation Problem Definition

Objective Function: Maximising Surface Coverage While Minimising Redundant Passes

The primary goal is to ensure comprehensive disinfection by maximising the surface area
covered by the robot, while simultaneously reducing unnecessary overlaps in its path. This
balance enhances efficiency and conserves energy.

Let:

 Atotal represent the total area requiring disinfection.

 Acovered denote the area effectively covered by the robot.

 Predundant signify the length of paths where the robot's disinfection overlaps.

The objective function J can be formulated as:

J=α
( A total)
A covered
−β Predundant

Here, α and β are weighting factors that prioritise coverage and redundancy minimisation,
respectively.
Constraints: Distance, Angle of Exposure, Time per Area

1. Distance Constraint: The efficacy of UV-C disinfection diminishes with increased

distance from the source due to the inverse square law. If I0I_0I0 is the initial intensity at
a reference distance d0, then the intensity I at distance d is:

2
d0
I ( d )=I 0 ( )
d

To ensure effective disinfection, I(d) must remain above a threshold Imin.

2. Angle of Exposure Constraint: The intensity of UV-C light also depends on the angle θ
between the light source and the surface. The effective intensity Iθ is:

I ( d )=I cos (¿ θ)¿

Surfaces must be oriented such cos(θ) remains sufficiently large to maintain Iθ ≥ Imin.

3. Time per Area Constraint: The disinfection dose D is a product of intensity and
exposure time t:

D=I ×t

To achieve a required dose Dreq, the exposure time must satisfy:

D req
t=
I

These constraints ensure that each surface receives adequate exposure for effective
disinfection.

4.2 Geometric and Kinematic Considerations

Robot Motion Modelling

Accurate modelling of the robot's movement is crucial for precise path planning. Assuming a
differential drive robot, its position (x,y) and orientation ϕ can be described by:

dx dϕ
=vcos (ϕ) =ω
dt dt
Here, v is the linear velocity, and ω is the angular velocity.

Path Curvature and Smoothness Constraints

To prevent mechanical strain and ensure consistent disinfection, the robot's path should avoid
sharp turns. The curvature κ of the path is given by:

ω
κ=
v

Imposing a maximum allowable curvature κmax:

|ωv|≤ κ max

This constraint ensures smooth and feasible trajectories.

4.3 Illumination Model

UV-C Intensity Function Over Distance

As previously discussed, UV-C intensity decreases with distance following the inverse square
law:

2
d0
I ( d )=I 0 ( )
d

Angle-Dependent Exposure Calculations

Considering the angle of incidence θ, the effective intensity on a surface is:

I ( d )=I cos (¿ θ)¿

Formulation of Exposure Coverage as a Mathematical Function

The total exposure E a point on a surface receives is the integral of intensity over time:

T
E=∫ I θ ( t ) dt
0

For a constant intensity during exposure time TTT:

E=I θ ×T

Ensuring E ≥ Dreq guarantees effective disinfection.

4.4 Graph Theory Representation of the Environment

Node-Based Representation for Path Planning

The environment is modelled as a graph G = (V , E), where:

 V represents discrete points (nodes) corresponding to specific locations or waypoints.

 E denotes edges connecting these nodes, representing possible paths.

Edge Cost Functions Based on Disinfection Efficiency

Each edge e ∈ E is assigned a cost c(e) that reflects the time and energy required for the robot to
traverse it, as well as the effectiveness of disinfection along that path. Factors influencing c(e)
include:

 Distance between nodes.

 Required exposure time to achieve Dreq.

 Overlap with previously disinfected areas.

5. Algorithmic Approach for Optimal Path Planning

In this chapter, we explore various algorithmic strategies to achieve optimal path planning for
UV-C disinfection robots. These strategies ensure that all surfaces receive adequate UV-C
exposure while maintaining efficiency and safety.

5.1 Modification of Existing Algorithms to Integrate UV-C Constraints

Traditional pathfinding algorithms like A* and Dijkstra's are designed to find the shortest path
between two points. However, for UV-C disinfection, it's crucial to modify these algorithms to
account for surface exposure requirements.

Adjusting A or Dijkstra’s for Surface Exposure

The standard A* algorithm uses heuristics to determine the shortest path. To adapt it for UV-C
disinfection, we can modify the heuristic to consider factors such as the intensity and duration of
UV-C exposure needed for effective disinfection. This ensures that the robot's path not only
covers all areas but also provides sufficient exposure time to deactivate microorganisms.

Similarly, Dijkstra's algorithm, which finds the shortest path in a graph, can be extended to
account for UV-C constraints. By incorporating surface exposure requirements into the
algorithm's cost function, the robot can plan paths that ensure comprehensive disinfection. For
instance, an extended Dijkstra algorithm has been proposed that models the environment using
Delaunay triangulation, allowing for smoother and more efficient paths on complex surfaces.

Hybrid Methods Incorporating Optimization Techniques

To enhance path planning, hybrid methods that combine traditional algorithms with optimization
techniques like Particle Swarm Optimization (PSO) and Genetic Algorithms (GAs) can be
employed. These methods optimize the robot's trajectory by considering multiple factors, such as
obstacle avoidance, energy consumption, and UV-C exposure levels.

For example, a study utilized bio-inspired algorithms to adjust a robot's coordinates and stopping
times at key points, ensuring complete disinfection coverage while minimizing the risk of
excessive UV-C exposure. By integrating PSO or GAs with pathfinding algorithms, the robot can
dynamically adapt its path to varying environments, achieving optimal disinfection efficiency.

5.2 Coverage Path Planning (CPP) Techniques

Coverage Path Planning focuses on creating paths that allow the robot to cover an entire area
systematically, ensuring no spots are missed during disinfection.

Exact and Heuristic-Based Solutions

Exact methods aim to provide complete coverage with mathematically proven strategies. These
methods often involve decomposing the area into smaller, manageable sections and planning
paths that cover each section efficiently. However, exact methods can be computationally
intensive, especially in complex environments.

Heuristic-based solutions, on the other hand, use rules of thumb or strategies that provide good
enough coverage with less computational effort. While they may not guarantee complete
coverage, they are often faster and more adaptable to dynamic environments.

Lawn-Mower Pattern vs. Spiral vs. Randomized Strategies

Several CPP strategies can be employed:

 Lawn-Mower Pattern: This method involves the robot moving back and forth in parallel
lines, similar to mowing a lawn. It's straightforward and ensures systematic coverage but
may be inefficient in irregularly shaped areas.

 Spiral Pattern: The robot starts at a central point and moves outward in a spiral. This
approach is effective in open areas and reduces the number of turns, which can save time.
  Randomized Strategies: The robot moves in random directions, which can be useful in
complex environments with many obstacles. However, this method doesn't guarantee
complete coverage and may require more time.

5.3 Real-Time Adaptive Path Adjustments Using Sensor Feedback

In dynamic environments, static path planning may not suffice due to unforeseen obstacles or
changes. Implementing real-time adaptive path adjustments allows the robot to respond to sensor
feedback, ensuring both safety and effective disinfection.

By equipping the robot with sensors that detect obstacles, surface types, and UV-C intensity, it
can adjust its path in real-time. For instance, if an obstacle is detected, the robot can recalculate
its route to avoid it while still covering all necessary areas. Additionally, sensors can monitor
UV-C exposure levels, allowing the robot to adjust its speed or distance from surfaces to ensure
optimal disinfection without causing damage.

Incorporating sensor feedback into the control system enables the robot to adapt to changes,
maintain efficiency, and ensure comprehensive coverage.

6. Comparative Analysis of Path Planning Approaches

Theoretical Comparison of Selected Algorithms

Path planning is essential for UV-C sanitization robots to navigate spaces efficiently and ensure
thorough disinfection. Several algorithms are commonly employed, each with unique
characteristics:

 A*: A grid-based search algorithm that finds the shortest path from a start to a goal point
by evaluating the cost of movement and the estimated cost to the goal.

 Rapidly-exploring Random Tree (RRT): A sampling-based algorithm that builds a tree

by randomly sampling points in the space, aiming to efficiently explore large, high-
dimensional areas.
  Probabilistic Roadmap (PRM): Another sampling-based method that creates a roadmap
of the space by randomly sampling points and connecting them, suitable for multi-query
path planning problems.

Strengths and Weaknesses in the Context of UV-C Sanitization

In the realm of UV-C sanitization, the effectiveness of these algorithms varies:

 A* Algorithm:

 Strengths: Provides optimal paths in known environments, ensuring complete

coverage.

 Weaknesses: Computationally intensive in large or dynamic spaces, which may

hinder real-time application.

 RRT:

 Strengths: Efficiently handles high-dimensional and dynamic environments,

quickly finding feasible paths.

 Weaknesses: Paths may be non-optimal and require smoothing; coverage may be

incomplete without modifications.

 PRM:

 Strengths: Effective in static environments with multiple queries, offering a

comprehensive map for repeated use.

 Weaknesses: Less suitable for dynamic settings; initial roadmap construction can
be time-consuming.

For UV-C sanitization, where complete and efficient coverage is crucial, algorithms like A* can
be enhanced. For instance, an improved A* algorithm has been shown to increase search
efficiency by 89.4% and reduce path length by 46.3%, optimizing disinfection processes.
Computational Complexity Analysis

Understanding the computational demands of these algorithms is vital:

 A*: Its time complexity is exponential in the worst case, especially in high-dimensional
spaces, making it less efficient as complexity grows.

 RRT: Generally faster due to random sampling, but the quality of the path depends on
the number of samples, which can affect performance.

 PRM: Initial construction has high computational costs, but once built, query resolution
is efficient.

7. Conclusion and Future Work

Summary of Key Findings from Literature and Mathematical Analysis

In this research, we have delved into the optimisation of path planning algorithms for UV-C
sanitisation robots, integrating insights from existing literature with our mathematical analyses.
Our primary findings are as follows:

 Enhanced Coverage Path Planning (CPP): Implementing advanced CPP algorithms,

such as the bio-inspired Glasius-based Neural Network (biGBNN), significantly
improves the robot's ability to navigate complex environments. This approach ensures
comprehensive surface coverage, effectively reducing the risk of missed spots during
disinfection.

 Integration of Artificial Potential Fields (APF): Incorporating APF into path planning
enhances obstacle avoidance capabilities. The APF-GFARRT* algorithm, for instance,
has demonstrated superior performance in dense environments, reducing pathfinding time
and optimising the robot's trajectory.
  Optimised Disinfection Efficiency: Strategic placement and movement of UV-C robots,
guided by precise path planning, can decrease disinfection time by approximately 30%
while increasing coverage by 35%, compared to static UV-C light placements.

Potential Areas for Future Research

While our study has advanced the understanding of path planning for UV-C sanitisation robots,
several avenues warrant further exploration:

 Human-Aware Path Planning: Developing algorithms that enable robots to operate

safely in environments shared with humans is crucial. Research into human-aware
coverage path planning can facilitate real-time adjustments, ensuring effective
disinfection without posing health risks.

 Adaptive Algorithms for Dynamic Environments: Investigating path planning methods

that allow robots to adapt to changing surroundings in real-time will enhance their
operational efficiency in varied settings.

 Standardisation of UV-C Disinfection Protocols: Establishing uniform guidelines and

regulations for the use of UV-C technology in healthcare settings is essential to maximise
its efficacy and safety.

Practical Implementation Challenges and Future Studies Needed

Translating theoretical advancements into practical applications presents several challenges:

 Shadowing Effects: Objects and complex geometries can create shadows, hindering UV-
C exposure to certain areas. Addressing this requires sophisticated path planning to
ensure all surfaces receive adequate irradiation.

 Safety Concerns: UV-C radiation poses health risks upon direct exposure. Developing
fail-safe mechanisms and operational protocols is imperative to protect both operators
and occupants.
  Environmental Variability: Diverse and dynamic environments, such as hospitals with
frequent human movement, necessitate adaptable path planning algorithms capable of
real-time responsiveness.

Addressing these challenges requires targeted research focusing on:

 Advanced Sensor Integration: Equipping robots with sensors to detect obstacles and
human presence can enhance safety and operational efficiency.

 Real-Time Path Adjustment Algorithms: Developing algorithms that allow for

immediate path recalculations in response to environmental changes will improve the
robot's adaptability.

 Comprehensive Safety Protocols: Establishing and adhering to strict safety guidelines

will mitigate health risks associated with UV-C exposure.

REFERENCES

1. Banjanovic-Mehmedovic, L., Karabegovic, I., Jahic, J., & Omercic, M. (2021). Optimal
path planning of a disinfection mobile robot against COVID-19 in a ROS-based research
platform. Advances in Production Engineering & Management, 16(4), 405–417.
https://doi.org/10.14743/apem2021.4.409
2. Conroy, J., Thierauf, C., Rule, P., Krause, E., Akitaya, H., Gonczi, A., Korman, M., &
Scheutz, M. (2021). Robot development and path planning for indoor ultraviolet light
disinfection. arXiv preprint arXiv:2104.02913. https://arxiv.org/abs/2104.02913
3. Correia Marques, J. M., Ramalingam, R., Pan, Z., & Hauser, K. (2021). Optimized
coverage planning for UV surface disinfection. arXiv preprint arXiv:2103.14137.
https://arxiv.org/abs/2103.14137
4. Mikhailovskiy, N., Sedunin, A., Perminov, S., Kalinov, I., & Tsetserukou, D. (2021).
UltraBot: Autonomous mobile robot for indoor UV-C disinfection with non-trivial shape
of disinfection zone. arXiv preprint arXiv:2108.09792. https://arxiv.org/abs/2108.09792
5. Perminov, S., Kalinov, I., & Tsetserukou, D. (2023). GHACPP: Genetic-based human-
aware coverage path planning algorithm for autonomous disinfection robot. arXiv
preprint arXiv:2307.08294. https://arxiv.org/abs/2307.08294
6. Pierson, A., Romanishin, J., Hansen, H., Yañez, L., & Rus, D. (2021). Designing and
deploying a mobile UVC disinfection robot. arXiv preprint arXiv:2104.02913.
https://sites.bu.edu/pierson/files/2021/11/pierson2021iros.pdf
7. Wang, S., Li, Y., Ding, G., Li, C., Zhao, Q., Sun, B., & Song, Q. (2022). Design of UVC
surface disinfection robot with coverage path planning using map-based approach at-the-
edge. Robotics, 11(6), 117. https://doi.org/10.3390/robotics11060117
8. Zhou, Y., & Tuzel, O. (2018). VoxelNet: End-to-end learning for point cloud based 3D
object detection. Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition, 4490–4499. https://doi.org/10.1109/CVPR.2018.00472
9. Redmon, J., & Farhadi, A. (2018). YOLOv3: An incremental improvement. arXiv
preprint arXiv:1804.02767. https://arxiv.org/abs/1804.02767
10. He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask R-CNN. Proceedings of the
IEEE International Conference on Computer Vision, 2961–2969.
https://doi.org/10.1109/ICCV.2017.322
11. Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster R-CNN: Towards real-time object
detection with region proposal networks. Advances in Neural Information Processing
Systems, 28, 91–99. https://arxiv.org/abs/1506.01497
12. Simonyan, K., & Zisserman, A. (2015). Very deep convolutional networks for large-scale
image recognition. arXiv preprint arXiv:1409.1556. https://arxiv.org/abs/1409.1556
13. Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke,
V., & Rabinovich, A. (2015). Going deeper with convolutions. Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition, 1–9.
https://doi.org/10.1109/CVPR.2015.7298594
14. Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep
convolutional neural networks. Advances in Neural Information Processing Systems, 25,
1097–1105. https://doi.org/10.1145/3065386
15. LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied
to document recognition. Proceedings of the IEEE, 86(11), 2278–2324.
https://doi.org/10.1109/5.726791