WALL CLIMBING ROBOT PROJECT

Abstract

As part of the Mechatronics course in the Master 1 Bruface program at the VUB and ULB universities, students were tasked with designing a robot capable of performing a predefined task. Group 6, composed of six students, had to develop a Window Cleaning Robot. The goal of the project was to create a robot capable of adhering to vertical glass surfaces and performing automated cleaning operations.

The outcome of this project was a functional prototype capable of generating sufficient vacuum suction to adhere to flat surfaces and maneuvering in a controlled manner. However, extensive tests on wet surfaces have not been conducted. Videos demonstrating the robot's functionality and adhesion mechanism are attached in the section 9.

This is the list of the different materials used:

Electronic Materials:

One Brushless motors (RS 2205)

One microcontroller (ESP 32)

Four Speed reduction gear DC motor

One driver (L298N)

One ESC

One LiPo Battery

One PS4 remote controller

Wires

Non electronic materials:

Propeller ( 6.5 cm of radius)

Four 3D printed wheels (4 cm)

Microfiber Cloth

Rubber

A laser cutted chassis (MDF, thickness:3mm)

1. (Table of contents)
2. Project Motivation
3. Functional Analysis/requirements
4. State of the Art and Patent Analysis
5. Conceptual Design
6. High-Level Design / Embodiment Design
7. Design of Sub-Systems
8. Integration Guide
9. Demo Project Show & Quick Start Guide
10. Critical Review of the Project
11. Sustainability
12. Bill of Materials
13. Presentation of the team
14. Project Repository

Problem to Solve

With the rapid urbanization of the modern world, glass has become a dominant architectural element in cityscapes, from residential homes to massive skyscrapers. However, maintaining the cleanliness of these surfaces presents significant challenges. The primary issue is safety. Window cleaning, particularly for high-rise buildings, is classified as a high-risk profession. Manual cleaning often requires the use of ladders, scaffolding, or rope access systems (abseiling), exposing workers to the risk of fatal falls and severe injuries(The latest accident reported in the media in Brussels September 4, 2025) .

Furthermore, the manual process is labor-intensive and costly. For large commercial buildings, the logistical complexity of deploying cleaning crews results in high maintenance costs and infrequent cleaning cycles. In the residential sector, many homeowners struggle to clean exterior windows on upper floors or difficult to reach skylights, often neglecting them due to the physical difficulty or the high cost of hiring professionals.

From an efficiency standpoint, dirty windows reduce natural light penetration, impacting the well-being of occupants and increasing the reliance on artificial lighting. In the specific case of solar panels, dust and debris accumulation can significantly reduce energy output.

A fully automated window cleaning robot addresses these critical issues. By replacing human labor in dangerous environments, it eliminates the risk of injury. Additionally, it offers a cost-effective, on-demand solution for maintaining the aesthetic and functional quality of glass surfaces, reducing the time and effort required for both private and public sectors.

Moreover, market analysis validates this growing need, with over 3.8 million units sold in 2023 and a projected growth rate of 9.6% through 2035, confirming a global shift towards automated maintenance solutions to minimize risk and labor [1].

Persona Identification

The window cleaning robot is designed to meet the specific needs of various groups and organizations responsible for building maintenance and efficiency. The main potential customers include:

1. Residential Homeowners: Individual homeowners, particularly those with multi-story houses or large bay windows, often lack the equipment to clean exterior glass safely. This group requires a consumer-friendly product that is affordable, lightweight, and easy to set up. Their primary goal is to maintain their home's appearance without climbing ladders or hiring expensive services.
2. Retail Businesses and Showrooms: For car dealerships, clothing stores, and restaurants, clear windows are essential for displaying products and inviting customers. These businesses need a robot that is agile enough to navigate framed glass doors or shop windows efficiently, ensuring a pristine look on a daily basis without disrupting business operations.
3. Solar Farm Operators: While not strictly "windows," the glass surfaces of solar panels share similar cleaning requirements. Operators of large solar arrays need automated solutions to remove dust and bird droppings that reduce energy efficiency. For this persona, the robot must be autonomous and capable of working on inclined surfaces to maximize the power plant's output.

References: [1] Market Growth Reports. (2025). Window Cleaning Robot Market Size | Global Analysis [2035]. Retrieved from https://www.marketgrowthreports.com/market-reports/window-cleaning-robot-market-103810

Use Case

The operational cycle of the robot is illustrated in the diagram provided. It begins with the Preparation (Block 0) phase, where the operator checks the battery level and the condition of the cleaning pads. Next is Transport & Setup (Block 1): the robot is placed on the glass surface, and the adhesion system is activated. Once securely attached, the Cleaning Process (Block 2) begins, where the robot moves across the window (at approximately 10 cm/s) under user control to perform the cleaning task. Finally, the cycle ends with Turn-Off & Storage (Block 3): the robot is safely removed after deactivating the suction and stored for future use.

Functional Analysis

Based on the usage scenario described above, we can derive the specific functional requirements necessary for the robot's operation. To enable the secure transition from setup to the cleaning phase, the primary requirement is the ability to adhere to vertical glass surfaces; specifically, the fan force mechanism must be robust enough to maintain a downforce greater than 1.5 times the weight of the robot to prevent falls (fig 3.4). Additionally, the operational cycle requires the robot to navigate against gravity, implying a need for a drive system capable of moving up, down, left, and right while overcoming the friction generated by the cleaning pads. As the core purpose of the cleaning block is hygiene, the robot must be capable of effectively removing dust and grime without leaving streaks or scratching the glass. Finally, because the usage scenario defines the navigation mode as Manual / Free path, the system necessitates a remote control (via Bluetooth) that allows the user to directly guide the robot's movements. The fully functional analysis is summarized on (fig 3.3).

Requirements list

Following the Functional Analysis, which established the robot's core capabilities and behavioral logic, these qualitative needs are translated into specific technical constraints and performance metrics. These specifications are prioritized into critical and operational categories to guide the engineering phase.

The Critical Requirements focus on the fundamental physics, safety, and commercial viability of the device. To guarantee the secure adhesion identified in the analysis, the suction system must generate a downforce greater than 1.5 times the robot's weight, requiring a minimum thrust of 9N. Simultaneously, the total mass must be strictly minimized to a maximum of 600g to ensure efficiency. Safety is further enforced by a mandatory failsafe mechanism, such as a backup rope, to mitigate the risk of falls. Functionally, the robot must achieve a cleaning coverage of over 90% per cycle. These technical goals operate within a strict economic constraint, requiring a Bill of Materials of less than 100 €.

The Operational & Environmental requirements refine the hardware specifications and user experience. The power architecture relies on wireless operation using a LiPo 3S (11.1V) battery to maintain the specified cleaning speed of approximately 10 cm/s. To withstand the working environment, the design requires an IP54-equivalent resistance to water and cleaning fluids, alongside a Velcro system for easy pad replacement. Finally, to ensure the device is suitable for a domestic setting, the acoustic noise level must be limited to less than 65 dB.

The mass criterion is derived from the bill of materials, which provided a range of BLDC motor power values. These values were used to determine the available thrust, and consequently to estimate the total system maximum mass using the equation presented in Figure 3.4.

This requirements list is available at (fig 3.2)

Energy Efficiency,Power Management & Eco-Design

The eco-design strategy for this project addresses three critical aspects: energy optimization, waste reduction, and lifecycle reparability. Regarding energy management, while the robot’s operation in direct sunlight initially suggests the integration of onboard solar panels, this solution proves impractical due to the high power demand of the vacuum motors required to maintain vertical adhesion against gravity. Consequently, the design prioritizes high-density rechargeable batteries to optimize the power-to-weight ratio, with the potential integration of a solar-coupled docking station to maintain renewable energy use without burdening the robot's mass. Simultaneously, to mitigate operational waste, the system rejects disposable cleaning wipes in favor of a Velcro-based attachment mechanism for washable and reusable microfiber pads, significantly lowering the environmental impact over the product's lifecycle. Finally, the mechanical design emphasizes reparability and sustainable materials; the prototype utilizes standard nuts and bolts instead of permanent adhesives to facilitate the replacement of critical components like motors or batteries, while the chassis employs laser-cut MDF, offering a more biodegradable and accessible alternative to complex injection-molded plastics during the prototyping phase.

Conclusion

In conclusion, the goal of this project is to develop a robot capable of adhering to and cleaning vertical glass surfaces while respecting the strict weight and safety constraints. However, during the development of our prototype, the scope of the prototype has been adjusted to focus on the reliability of the adhesion and locomotion mechanisms under manual remote control. This proof of principle prioritizes the validation of the physical constraints over complex software navigation. Regarding the docking station and autonomous charging, these features were descoped from the final prototype. Implementing a safe charging circuit for LiPo batteries within the project's timeframe and safety regulations proved unfeasible; therefore, external charging is required. The final solution thus balances the critical need for safety and cleaning efficiency with the practical limitations of a prototyping environment.

4.1 State of the art:

This shows an overview on wall climbing robots already on the market. And compare them with the group's prototype and see if some meets already the criterias the group defined.

Below are listed the most popular wall cleaning robots and a description of each:

- Ecovacs winbot w1: It's a beautiful white robot with a button "power" at the front of his upper surface.There are two nozzles at the front and behind the robot who are responsible of the spray of the cleaning liquid.the robot has two tracks for his movement. At the center, there is a suction sysytem to hold the robot against the wall, a cleaning fluid level indicator at the bottom right.

There are two cables behind the robot, one for the power supply and one for the safety system.Ecovacs is equipped with a clever storage case.It has 2 types of sensors, the first act like a bumper.Placed on each corner, it will cushion any impact against the window's edge.The second sensor is an anti-glare edge detection technology,useful to clean frameless windows.It costs around 450 euros.

-Mamibot W-120T: It's useful for indoor and outdoor purposes. It weights 1.35 kg and is a squared good to reach the edges. It's a remote controlled robot, the robot is equiped with an AI technology.It has a battery of 650 mAh as power supply.Concerning the safety, Mamibot has an anti-fall sensor and a sturdy safety rope.To move, the robot uses tracks.It cost around 250 euros on amazon.

- HOBOT 388: It has a built in tank and automatic cleaning solution spray, it's an ultrasonic diffuser system.For the cleaning part,the robot has rotating microfiber pads which scrub the glass surface and remove dust microparticules by electrification.

With the Navig+TM navigation technology you can put the robot anywhere,it will automatically go at the top of the window to start the work. It's a remotely controllable with a connexion bluetooth with a smartphone. Wheels are responsible of the movement.

It has a 4m extension cord for power supply and a rope system for security; also a backup battery. It costs around 380 euros on amazon.

4.2 Patent analysis:

The objective is to determine whether certain parts of the project are intellectually

protected by existing patents. If a specific technical solution is protected by one or more patents, it cannot

be directly integrated into the design. Conversely, in the absence of relevant patent protection, a freedom

to operate exists, meaning that the design is not legally constrained.

This analysis was conducted using the European Patent Office (EPO) database. Several keyword searches

were performed in order to identify relevant intellectual property related to the field of interest. The selected

keywords were:

• Wall climbing robot

• Cleaning robot

• Window cleaning robot

This is a detailed list of the analyzed patents:

• CN210235142U

• US6793026B1

• WO2025119408A2

• US2015272411A1

• US2015282684A1

• US10265734B2

• CA2970696A1

These documents mainly describe

specific mechanical architectures, adhesion systems, or control strategies for robotic cleaning or wall-climbing

devices.

No identified patent was found to directly correspond to the technical solution developed in this project.

In addition to the keyword-based patent search, an analysis of existing competitors and commercial prod-

ucts was also carried out using the European Patent Office database. This complementary study aimed to

identify patents and technologies associated with well-known manufacturers already present on the mar-

ket, such as Ecovacs and other competing companies.

Based on both the European patent search and the competitor analysis performed through the EPO

database, no patent or market solution was identified that directly blocks the development of the studied

robotic concept. The concept therefore remains distinct from existing products while addressing a similar

application domain.

Conceptual Design

5.1 Problems and Means

In order for the robot to be as efficient as possible, it must overcome a number of problems often encountered with this type of robot.

This is clearly illustrated in the following which also provides the appropriate general solution for each problem:

5.2 Comparative Analysis of Technical Means

Before formulating complete concepts, we analyzed the specific components available in our Morphological Chart to determine the most efficient building blocks.

1. Adhesion to Glass

Propeller with BLDC Motor: This active adhesion method uses a high-speed turbine to create a massive

pressure difference, pressing the robot against the glass. Unlike passive cups, it allows the robot to cross

non-airtight areas without falling. However, it requires constant high power and generates significant noise.

Suction Cups: Passive suction cups provide a silent and strong hold. However, they require a perfectly

smooth surface; a single crack or grout line breaks the vacuum seal, causing the robot to fall. The movement

is also discontinuous and slow, as the robot must seal and unseal sequentially to step.

Gecko Adhesive: Inspired by lizard feet, this dry adhesive uses Van der Waals forces. It is silent and

energy-free for adhesion. However, the technology is expensive and highly sensitive to contamination; once

the pads get dusty, the adhesion force drops drastically.

1. Movement System

Rotating Disks: This method combines movement and cleaning (the cleaning pads are the wheels).

While this minimizes part count, it results in a "waddling" motion that is difficult to control precisely. It

inevitably leaves circular swirl marks on the glass.

Caterpillar Tracks: Tracks provide excellent surface area and traction. However, they introduce high me-

chanical friction, especially when turning (skid-steering), which drains the battery. Furthermore, a track

system is mechanically complex and heavy, making it difficult to stay under the strict 600g weight limit.

Rubber Wheels: Standard rubber wheels offer high traction, low rolling resistance, and precise linear

control. They are lightweight and allow the robot to drive in straight lines a critical requirement for effec-

tive squeegee usage.

1. Cleaning Mechanism

Squeegee + Spraying: This mimics professional manual cleaning. The fluid dissolves dirt, and the squeegee

mechanically removes the dirty water without leaving streaks. It requires precise motion control but offers

the best visual finish.

Wipe + Spraying: Common in rotary robots. Microfiber pads wipe the glass, but if the window is very

dirty, the pads quickly saturate and end up spreading the dirt rather than removing it.

Vibrating Pad: A pad oscillates at high frequency to scrub stuck-on dirt. While mechanically effective

for scrubbing, it adds weight and vibration that can confuse sensors. It also lacks a mechanism to actually

remove the dirty water from the glass (Technology heavily patented (Hobot), limits commercial implemen-

tation).

Suction: Vacuuming is effective for loose dry dust but completely ineffective against stuck-on grime, rain

spots, or bird droppings. It does not "wash" the window.

1. Energy Supply

Cable + Small battery : This hybrid approach uses a power cord as the primary energy source, backed by

a small emergency battery. It provides continuous high power to the vacuum motor without the weight

penalty of a large battery pack, making the target weight achievable. The onboard battery acts solely as an

Uninterruptible Power Supply (UPS), maintaining adhesion for 20 minutes during a power outage to ensure

safety.

Li-ion Battery: The industry standard. It offers high energy density and is capable of delivering the high

current bursts required by the BLDC vacuum motor and traction motors. It keeps the robot wireless and

compact.

Solar Energy Panels: While eco-friendly, this is impractical for the primary power source. The surface

area available on a compact robot cannot generate sufficient wattage to run a vacuum pump and motors

simultaneously.

1. Control Navigation

Infrared Sensors: These are lightweight, fast, and cost-effective. They are reliable for detecting "cliffs"

(frameless window edges) and measuring short distances. Unlike some optical sensors, they work reason-

ably well on glass surfaces.

Ultrasonic Sensors: Robust for general obstacle detection, but unreliable on glass. Sound waves can de-

flect unpredictably off hard, smooth surfaces at angles, leading to inaccurate distance readings.

AI Camera: Allows for dirt detection and smart pathing. However, it requires a heavy, power-hungry pro-

cessor to analyze video feeds in real-time, which is overkill for a simple coverage pattern.

LIDAR Camera: Provides perfect mapping. However, standard LIDAR lasers often pass through clear

glass rather than detecting it, and the hardware is too bulky and expensive for a low-cost, lightweight robot.

1. Data Management & Algorithms

Microcontrollers and Feedback Loop: A centralized architecture allows for real-time control with mini-

mal latency. This is crucial for safety: if the pressure sensor detects a vacuum drop, the MCU must adjust

the motor speed in milliseconds.

Communication Bus: A distributed architecture (like CAN bus). While useful for complex modular sys-

tems (like cars), it introduces unnecessary wiring and protocol complexity for a small robot with a limited

number of actuators.

1. Detachment & Retrieval

Handle: The simplest, most fail-safe method. A physical handle allows the user to securely grip the robot

before turning off the suction. It adds negligible weight and never malfunctions.

Docking Station: Allows for autonomous charging. However, mechanically aligning a vertical robot with

a charger is complex. It adds weight to the robot and requires permanently mounting a station on the win-

dow, which is visually intrusive.

1. Safety System

Security Switch: An active safety measure. This sensor monitors vacuum pressure or wheel contact. If it

detects a failure, it triggers an emergency protocol to prevent a fall before it happens.

Parachute System: Physically impractical. A parachute requires significant height to deploy and slow a

fall. For a residential window, the robot would hit the ground before the parachute could open.

Rope System: A passive backup. While essential for ultimate safety, it is a passive tether. The Security

Switch is preferred as the primary system because it actively attempts to prevent the fall from occurring in

the first place.

5.3 Design

In order to have the most complete solution possible (resolving most of the problems presented earlier),

the group decided to choose between three different concepts. Each concept is a unique combination of

the proposed means from the morphological chart.

To ensure a fair comparison, all three designs share a common architecture: Fan-based Adhesion (BLDC)

to guarantee strong holding force, a Microcontroller for logic, Infrared Sensorsfor edge detec-

tion, and a Security Switch.

The variations lie in the movement system, the cleaning mechanism, and the power strategy.

Concept 1: The "Heavy Duty" Tracker

This concept focuses on maximum traction and a fully wireless experience.

The choices made for this robot are shown in (Fig. 5.1).

Concept 2: The "Lightweight" Wiper

A simplified version focusing on weight reduction.

The choices made for this robot are shown in (Fig. 5.2).

Concept 3: The "Squeegee Pro" (Selected Design)

This concept aims to mimic professional manual cleaning. It combines precise linear movement with

the most effective cleaning mechanism and optimal power-to-weight ratio.

The choices made for this robot are shown in (Fig. 5.3).

5.4 Weighted Decision Matrix

To select the final design, we scored each concept based on the importance of the features defined in the

Functional Specifications. Scoring Rule: The maximum score for each criterion corresponds to its Impor-

tance level (e.g., if Importance is 5, a perfect solution gets 5, a poor one gets 1 or 2)

The results are present on (fig 5.4)

5.5 Decision

• Concept 1 (Score: 25/41): This concept is penalized heavily on the "Lightweight" and "Cost" criteria

(Importance 4 and 3) due to the heavy battery system. It also scores poorly on cleaning quality.

• Concept 2 (Score: 37/41): A strong contender. It scores maximum points on weight and agility. How-

ever, it loses critical points on the most important feature: "Clean the Surface" (Importance 5), as the

wiping pad is prone to smearing.

• Concept 3 (Score: 40/41): The clear winner. It matches Concept 2’s agility and lightweight advantages

but achieves a perfect score (5/5) in "Cleaning Quality" thanks to the Squeegee integration. It provides

the best compromise between performance, weight, and cost.

5.6 Prototype and Objective

However, during the development of the project, several design choices were made that slightly diverged from the initial objective, such as the use of a self-contained battery system. These decisions were nonetheless essential, as they significantly facilitated access to the required materials, notably through FablabStock. Furthermore, the chassis was manufactured in MDF, leveraging the resources of the fablab, and the final wheels of the prototype were 3D printed after an original wheel broke shortly before the deadline. Moreover since we didn't try on wet surfaces for now, we didn't put the squegees on but only the wipes. Overall, these choices allowed the project to be completed within the given constraints while maintaining a functional and coherent prototype.

High-Level Control Architecture

Before detailing the mechanical embodiment, it is essential to define the logical behavior of the robot(fig 6.1) illustrates the main software control loop implemented in the microcontroller.

The sequence begins with System Initialization, where the hardware peripherals (Motor Drivers, Bluetooth module, Sensors) are calibrated. The robot then enters a standby mode, waiting for a Bluetooth Connection to be established with the remote interface.

Once connected, the system enters a main loop that processes incoming data packages based on two primary functions:

1. Motion Control (Left Branch): When a navigation command is received, the algorithm parses the directional input and updates the Motor Drivers (PWM signals) to move the robot's wheels.
2. Status Monitoring (Right Branch): The system periodically checks critical safety parameters. As shown in the diagram, it monitors internal state variables, specifically checking for a "stop and pick-up" request. If this flag is raised, data is sent to trigger a warning or an emergency stop. Moreover, the fan speed is controlled periodically via the L2 and R2 buttons of the PS4 controller.

(Fig 6.1) Control Logic Flowchart

CAD Model description

Based on the previous studies, a complete CAD Model of the robot was created. This model includes all the main functional components as well as additional mechanical parts required for practical assembly, safety, and structural integrity.

Overall outside overview

The figures shows the complete robot from the outside.

On the top, it is possible to see:

1. a handle, used to easily carry the robot;
2. an ON/OFF button;
3. ventilation holes for the fan, designed to allow airflow while ensuring user safety.

On the side, there is:

1. a charging port, allowing the robot to be recharged safely.

On the bottom, there is:

1. two support bars, designed so that the cleaning cloth can be placed on them and removed easily when needed.

Internal Structure

The figure shows the robot from a more internal point of view.

The robot is composed of three main structural parts:

1. Bottom housing
2. Middle plate (component support structure)
3. Top housing

The bottom part of the robot includes raised supports.

The middle plate rests on these supports, which ensures proper positioning. Those support grows higher and thinner so that the bottom part fit into hols in the middle plate, ensuring correct alignement. These parts are then secured using bolts.

For the connection between the middle part and the top housing, cantilever snap-fit mechanisms are used as we can see. This solution allows a secure fixation, no visible screws on the outside and a clean, professional finish.

Fixation of internal components

The figure shows our components are fixed on the middle plate.

Components that include mounting holes are fixed using bolts and screws. Other components, such as the ESC (Electronic Speed Controller) and the battery, are placed inside deidcated holders cause they do not have holes initially. Those holders include mounting holes so they can be securely fixed using screws and bolts.

For the wheel and the motor, their assembly are a little more complex.

The motor is fixed to a dedicated support plate. A mechanical structure is placed between the motor and the wheel to ensure proper alignement and a clamping screw is used to lock the assembly and prevent any relative movement.

Material selection analysis

Voici le texte complet, où j'ai remplacé les tableaux par des listes ou des phrases intégrées pour que la lecture soit fluide.

At this stage, the materials used in the robot must be selected. This selection is performed using the Granta Edupack software. It concerns the external housing, the middle plate used to support the electronic components, and the enclosures for the battery and the ESC. Although the external housing is directly exposed to the environment, all these parts belong to the robot structure and share similar constraints. For this reason, the same material limits are applied in this study.

Before performing the material selection, it is necessary to define the function, objectives, and constraints of the robot structure.

The robot is designed to adhere to a glass surface and clean it, both on the inside and the outside. Therefore, the structure must be lightweight in order to not increase the load for the adhesion system, while remaining rigid enough to support its own weight and the electronic components.

In addition, the structure must withstand mechanical stresses such as vibration during the operation and possible impacts against the glass surfaces. The robot also contains electronic components, so the materials must also provide electrical insulation to ensure safety.

Based on these considerations, the selection criteria are defined as follows:

1. Function: Maintain adhesion to the glass surface during cleaning.
2. Objective: Minimize the mass of the structure while ensuring sufficient rigidity to support mechanical loads and resist deformation over time.
3. Constraints: The structure requires low weight, sufficient stiffness, resistance to vibrations and impacts, electrical insulation, and affordability.
4. Material Variables: Density, Young's modulus, Price, Electrical resistivity, and Fracture toughness.

As shown in Figure GrantaEdupack1, based on the defined objectives, density and Young's modulus are selected as the axes of the chart, in order to identify materials that provide a good compromise between low weight and high stiffness.

At the beginning of the selection process, all materials available in the database are considered. However, based on the defined constraints, the number of possible materials can be significantly reduced. Those constraints are implemented in the limitation module of the software and explained below.

First, the density is limited to a maximum value of 1500 kg/m³. This limit ensures that the robot remains lightweight. For example, a plate with dimensions of approximately 20 cm × 20 cm × 3 mm would have a mass of about 180 g at this density. This value is relatively high but remains acceptable for the robot structure.

The price of the material is limited to 5 euro/kg in order to keep the overall cost of the robot acceptable. Since the electronic components are already relatively expensive, the material used for the structure must remain affordable to ensure that the final product can be accessible to users.

The Young’s modulus is set to a minimum value of 1 GPa to guarantee sufficient rigidity. With such a value, the material does not deform significantly under the robot’s own weight. Even under external loads of the order of 100N, the resulting deformation remains very small and acceptable for the intended application.

To avoid brittle behavior, a constraint is also applied on the fracture toughness, which must be greater than 2 MPa·√m. This ensures that the material can resist impacts and does not fail suddenly by cracking.

Finally, the material must be a good electrical insulator. This requirement is essential to protect both the electronic components and the user, especially because the robot operates in a humid environment.

After applying all these constraints, many materials are eliminated from the selection process. The remaining candidates mainly belong to the thermoplastic polymer family as shown in Figure Granta2.

In order to select the best material among the remaining candidates, the initial objective must be recalled. The aim is to obtain a structure that maximizes stiffness while minimizing mass. For this reason, the Young’s modulus must be maximized while the density is minimized, leading to the definition of the following performance index:

M = E / ρ

Where E is the Young's modulus and ρ is the density.

Since the diagram uses logarithmic scales for both axes, the performance index must be expressed in logarithmic form:

log(E) = log( ρ) + log(M)

This equation represents a straight line in the log–log diagram of Young’s modulus versus density, with a slope equal to unity. By introducing this condition in the Granta EduPack software and moving the performance line upward, the materials with the highest values of the performance index can be identified.

As shown in Figure Granta3, ABS (Acrylonitrile Butadiene Styrene) is identified as the optimal material. ABS is a thermoplastic polymer. It offers a good compromise between low density and sufficient rigidity, while also providing good impact resistance and electrical insulation. In addition, ABS is inexpensive and easy to manufacture, which makes it well suited for the intended application.

In commercial applications, ABS is a widely used commercial thermoplastic material. It is commonly found in robot housings, electronic enclosures, and consumer products.

Other commercial materials, such as polycarbonate (PC), are widely used and provide higher impact resistance and slightly higher stiffness. However, PC is more expensive and is generally used for applications requiring very high mechanical resistance.

Manufacturing process

Once the material selection has been completed, the manufacturing process can be selected.

To determine which process to choose, an analysis must be performed: the robot is designed to be a commercial product, which implies medium to large production volumes. As said in the CAD section, the structure is composed of several separate components that are assembled. Each component has a specific geometry, including snap-fit features, holes for bolts, and complex three-dimensional shapes.

Shaping

The first step of the selection is the shaping stage. The objective of this step is to identify manufacturing processes capable of producing the required geometries.

Using Granta Edupack, shaping constraints were applied by selecting thermoplastics as the material family and 3D solid as the component geometry. Additional limits were also introduced in the software:

1. Mass range: 0.1 – 0.5 kg
2. Range of section thickness: 1 – 5 mm
3. Economic batch size: Minimum 1000 units

After applying these constraints, several manufacturing processes remain technically feasible, such as compression molding, drilling, injection molding, and machining.

Process comparison and selection

Thus, a comparison between the remaining processes is performed.

Drilling and machining are eliminated from the selection because they are not primary manufacturing processes that suit the features required in the design (such as snap-fits).

Compression molding is a possible solution but it is less adapted than injection molding to complex details.

So injection molding clearly emerges as the most suitable manufacturing process. It is fully compatible with ABS, allows the integration of snap-fit features and holes. It provides a good surface quality and ensures excellent repeatability. Injection molding is also cost-effective for medium or large production volumes, which is good for this project.

Mechanical subsystem

As stated in the previous section, the main objective is to maintain good adhesion to the glass while performing the cleaning task. To clearly define what the robot must be capable of and which aspects need to be implemented, a list of requirements was established in Section 3. In Section 5, a conceptual design was developed to determine the key features of the robot. Finally, Section 6 presents the embodiment design, which specifies the materials to be used and the exact placement of each component.

The previous sections focused on the final commercial product. However, during the development of the prototype, several challenges were encountered, which required specific design choices to ensure proper operation.

For the prototype, due to the limited available power, the robot was designed to be as lightweight as possible. To reduce the onboard mass while operating on the window, the robot was controlled using a remote controller. In order to further minimize weight and limit the number of onboard components, the battery was carried externally by one of the team members.

Several design requirements were therefore defined for the prototype. It had to be lightweight, capable of accommodating the wheels, the motor driver, and the ESP32, and ensure proper cable management so that the wires do not interfere with the fan or become disorganized.

Regarding the materials used for the prototype, the chassis was manufactured from 3 mm thick MDF and cut using a laser cutter. This thickness was selected as an optimal compromise between structural robustness and low weight. A large central opening was made in the chassis to accommodate the propeller and its motor. The motor was mounted using a dedicated support fixed to the chassis with screws and bolts.

The motor driver and the ESP32 were assigned specific locations on the chassis, and mounting holes were created to secure them with screws. The wheels were manufactured using 3D printing and covered with a rubber layer to increase friction. The wheels had to be custom-made due to last-minute issues with the originally available wheels, which were not correctly sized to ensure proper torque transmission from the wheel motors. While the original wheels worked adequately on the ground, they failed to provide sufficient traction on the glass surface, where higher friction and the suction force generated by the fan required improved torque transmission.

To solve this issue, custom 3D-printed wheels were designed and implemented. The wheels were fixed to the chassis using cable ties, and wooden shims were inserted between the wheels and the chassis to position them as close as possible to the window surface, thereby maximizing adhesion.

To solve this issue, custom 3D-printed wheels were designed and implemented. The wheels were fixed to the chassis using cable ties, and wooden shims were inserted between the wheels and the chassis to position them as close as possible to the window surface, thereby maximizing adhesion.

Regarding the wiring, cables of appropriate lengths were selected to minimize excess wiring. Two holes were added on the sides of the robot to improve cable routing and organization. In addition, protective cable sleeves were used to bundle wires that originate from and lead to the same locations, ensuring neat cable management and preventing interference with moving components.

Circuitry and sensors

The initial goal of our project was to design an autonomous robot capable of spraying water on a vertical glass surface. However, due to adhesion issues on the window, we had to revise some design choices and adapt the system.

As a result, the project was divided into two parts:

1. one part integrated directly on the robot,
2. and another part working separately

This section focuses on the electronic components used on the robot prototype and explains why each component was chosen and the figures shows how they are implemented.

Components on the prototype:

ESP32 Microcontroller

At first, we worked with an Arduino Uno. However, we decided to switch to an ESP32.

The main reason is that the ESP32 has built-in Bluetooth, which allows us to control the robot wirelessly using a PS4 controller. This avoids the need for an external Bluetooth module.

In addition, the ESP32 is compact and lightweight.

Gear Motor with Reduction

The robot uses gear motors with reduction to drive the wheels.

These motors are:

1. small and lightweight,
2. capable of producing a high torque,
3. easy to control.

A high torque is essential because the robot is pressed against a vertical glass surface. The wheels must generate enough friction to move the robot, which requires more torque than a normal horizontal motion.

Therefore, using gear motors with reduction is necessary to ensure reliable movement on the glass.

They are used with a power supply.

L298N Motor Driver

To control the wheel motors, we use an L298N motor driver.

This driver is based on an H-bridge.

It allows:

1. control of the rotation speed using PWM,
2. control of the rotation direction (forward and backward).

We can only use one cause this driver has been created to control to modul at the same time. It is a good point for our robot because we have four wheels, and wheels on the same side can be controlled together. So we can use one, and that's a good point for the robot to be as light as possible.

LiPo Battery

The robot requires a high power to operate the motors and the fan.

A laboratory power supply is not suitable because it cannot provide large current. A LiPo battery, on the other hand:

1. provides a stable DC voltage,
2. can deliver very high currents instantly,

For these reasons, a LiPo battery is essential for this project.

ESC (Electronic Speed Controller)

The ESC is a mandatory component when using a brushless motor.

The battery provides DC current, while a brushless motor (used for the fan) requires a three-phase alternating current. The ESC performs two essential functions:

1. it converts DC voltage from the battery into a three-phase signal,
2. it controls the motor speed.

Without the ESC, the brushless motor cannot operate.

RS2205 Brushless Motor (Fan Motor)

We chose an RS2205 brushless motor to drive the fan used for adhesion.

This motor is:

1. lightweight,
2. capable of delivering high power,
3. designed for high rotational speed,
4. well adapted for propellers.

The fan produces a horizontal thrust that presses the robot against the glass. This normal force increases the friction between the wheels and the glass, allowing the robot to move vertically.

We can compute the thrust needed for the robot:

If we assume an electrical power of 278W (11.1V and 25A) provided by the battery and if we consider having some loss and having a global efficiency of 70%, the mechanical power is 195W.

The thrust generated by the fan is given by :

T=(P_mech * (​2*ρ*A)^(1/2)​)^(2/3) ~10,8N

where :

1. ρ is the air density (1.225 kg/m^3)
2. A=pi*r^2 is the propeller of the disk area (r=0.065m)

We know that the friction force(F_frict) equals μ*T, where μ is the friction coefficient of the rubber on the glass (~0.6). So F_frict = 6.5N

To have our robot adhering on the window, we need: F_frict>mg. So, m, the mass of the robot, needs to be:

m<(6.5/9.81)~0.66kg

Components not on the prototype

VL53L0X Distance Sensors

These sensors are:

1. very small and lightweight,
2. accurate (millimeter precision),
3. based on a laser beam,
4. unctional in daylight and indoor lighting.

They measure distance along a narrow cone, not a full surface. This makes them ideal for controlling the distance between the robot and the glass surface.

12V water pump

For spraying water, it is attache to two pipes, one connected to the recipient and the other one is close at the end and have several little holes to have the effect of spraying water

5V Relay

The pump functionning with 12V, it is used as an eectrical switch to safely control the pump.

Software

On the ESP32, we implemented a program that controls the behavior of the robot according to the project requirements.

The overall operation of the robot is described in the flowchart diagram.

Using a PS4 controller, the user can command the robot’s movements and adjust the fan speed in real time.

This allows precise control of both the robot motion and the adhesion force generated by the fan.

We can't upload code either in ino and txt file it will be in the zip

The following is a brief list of instructions to set-up the system and make it work properly.

1. Prepare the wheels: glue together the 3D-printed body of the wheel with the rubber tread; then glue the shaft of each of the four motors (Motoreducteur N20) inside the wheel bore in the wheel, aligning the shaft key with the wheel keyway
2. Fix the plastic fan together with the RS2205 brushless motor by using a nut; then fix the back of the motor to the laser-cut support beam: be careful not to put too much torque in the fixing, otherwise the fan will not spin. Manually check if it can spin properly.
3. Position and fix the driver and the microcontroller onto the wooden laser-cut chassis, using the screws and the nuts
4. Fix the four motors onto the specific spaces of the chassis by using two plastic ties each. Be careful to position the two ties as far as possible surrounding the metal case of the motor to prevent it from pivoting, and not to interfere neither with the wires nor with the gear box
5. Properly wire the system as shown in the electronic schematic. Solder the power wires of the four motors and let free the cable of the motor of the fan: it will be connected to the Li-Po battery. In order to manage the wires, gather them with spirally-cut rubber tubes and use the slots in the chassis.
6. Fix the beam with the fan together with the chassis, keeping the wires on the side and stuck underneath. To end, ensure the wires with plastic ties, especially the ones surrounding the fan, in order to prevent them from colliding with the fan.

→ Show photos and videos. Make a guided voice over

In the current project, several compromises were made to achieve a functional final version of the robot. As anticipated, the main challenges concerned reducing the overall weight, providing sufficient thrust, increasing friction on the glass surface, and consequently ensuring stable driving control.

The initial idea, which had to be abandoned but could be implemented in future developments, was to make the robot self-driving. A system based on three time-of-flight sensors was designed: the robot was intended to drive straight by measuring the distance of one side from the edge of the window and adjusting its driving direction based on the time derivative of this distance. This method was tested and resulted in unstable but acceptable motion, as it relies on the derivative of the measured distance and on the non-linear variation of the distance itself. (IMU sensors were not used because of cumulative error due to the integration of the angular acceleration, without additional absolute measurements, such as vision or beacons.)

Turning maneuvers were intended to occur whenever the front distance sensor detected a value below a predefined threshold. The turning strategy relied on a fixed time interval, empirically determined, assumed to be sufficient to perform a 180° pivot turn. However, this approach could not be properly validated, which ultimately led to abandoning the self-driving strategy in the final implementation.

Another feature that has been given up in the making, due to an excess of weight, is an active cleaning system. It was meant to be pumping water intermittently, with a 12V pump controlled by the microcontroller through a relay, from a water container to a cloth, using a rubber pipe.

Any weight increment must be followed by an increment of thrust that ensures the robot to be stuck on the wall. This means either a higher power needed from the fan system or a suction chamber that makes the sticking way more efficient. In our experience, we attempted to use a suction chamber made out of tape, but the friction was too high for the wheels to have enough torque to keep the motion. Thus, we ended up removing the suction chamber and keeping the chassis and the fan as close as possible to the wall. This compromise between the increment of thrust and the higher need of torque from the wheels must be taken into account.

In our prototype, the 12 V battery that supplies the fan motor is handheld to avoid increasing the onboard weight, while the four drive motors are powered from an external source. The 4 motors are controlled with a single L298N driver: on each side, the two motors are connected in parallel to the same H‑bridge channel, sharing the maximum current that the driver can provide. To increase the torque available at each side, a possible enhancement is to use two separate L298N drivers, so that each side is driven by its own full driver and can exploit the maximum allowable current at that voltage.

A problem we encountered with the wheels concerned power transmission and friction. Initially, we tested commercially available toy wheels. In addition to their low friction, the main issue was that the wheel bore did not fit properly onto the motor shaft. Furthermore, the plastic material was not sufficiently resistant: under load, the bore was damaged, resulting in poor torque transmission. Due to time constraints, we opted for custom 3D-printed wheels made of a more resistant plastic. This solution also allowed us to design a bore that fits the motor shaft much more accurately, ensuring reliable torque transmission.

A final improvement could definitely be a safety system, both physical, with a grid to prevent objects from colliding with the fan and a good redundancy of thrust, or software, like a system that detects when the battery percentage is low and brings the robot autonomously to the ground, for instance.

In conclusion, several improvements can be made that were not possible due to time constraints and due to the need of making numerous attempts before getting to the working prototype. They mainly involve safety, autonomous driving, and increment of power both in the adhesion system and in the motors of the wheels.

Even though this robot is already pretty eco friendly because it's made of Wood(MDF 3mm), with the used of washable microfiber cloth wish allows the same cloth to be reused hundreds of times, eliminating the need to create trash after every cleaning session. Furthermore, the cloth works effectively with just water or mild biodegradable soap, meaning the robot cleans windows perfectly without requiring harsh chemicals or generating unnecessary waste, We think there are some really cool ways to make it even better for the environment. The suction fan uses a ton of electricity to stick to the wall. If we add a skirt around the bottom for better sealing against the glass wall . This means we do not have to run the motor at full speed to keep it attached. It saves energy and keeps the battery healthy for a lot longer. Another way to manage power efficiently, we could implement a PID control loop which dynamically adjust the RS2205 suction motor’s speed via PWM, drawing only the necessary current to maintain adhesion by doing this we have programmed the robot to be smart about how it uses energy. Instead of running the suction motor at full power all the time, the robot automatically adjusts the fan speed to use only the exact amount of power needed to stick to the wall. This prevents wasted electricity and makes the battery last much longer Also, the wood is not very strong, especially if it gets wet from cleaning . To stop it from getting mushy or breaking, we could paint it with some watery glue to make it tough. This way, the robot body lasts for years, and we do not have to keep wasting materials to build new ones. Another thing is creating a mobile app to replace the PS4 Controller.

This materials list outlines all necessary components for our project, with an emphasis on functionality and cost-efficiency. The inclusion of essential components like high-torque gear motors, an ESP32 development board, and structural materials like PLA filament and MDF ensures the successful construction of the Wall Climbing Robot for Cleaning Glass Walls and Windows. The Wall Climbing Robot project requires a total initial investment of €150.50, which includes purchasing bulk supplies of PLA filament and MDF wood that are sufficient for multiple builds. A breakdown of expenses reveals that 45% of the budget is allocated to electronics specifically the PS4 controller and LiPo battery while the use of standard motors helps keep mechanical expenses down. Since 3D printing is used exclusively for the small wheels and the main structure is made from affordable wood, the actual material used per robot is very cheap; consequently, once the bulk materials are bought, the cost to build a second unit drops significantly to just €68.00.

Jules Haut: I completed my bachelor's degree in civil engineering at ULB. For this project, I focused on the mechatronic integration of the robot. My work involved selecting and ordering components, designing parts in CAD for 3D printing, and handling the wiring and final assembly. I also developed the code to interface the PS4 controller with the ESP32. My favorite part was the troubleshooting phase; it was very rewarding to overcome the technical challenges and see the hardware and software successfully work together.

Gaelanne Bouteille: I completed my bachelor's degree in civil engineering at ULB. My main contribution to the project was related to the conceptual design and mechanical development of the robot. I worked on early design concepts and mechanical architecture choices and contributed to CAD modelling and design justification. My favorite part of the project was developing and comparing different design concepts, as it allowed creativity while applying engineering constraints.

Selim Khalil: I completed my bachelor's degree in civil engineering at ULB. In this project, I focused on the state of the art, patent analysis, and requirements definition, ensuring that the proposed solution was relevant, feasible, and innovative. I also contributed to the functional analysis and helped translate needs into technical constraints. My favorite part of the project was analyzing existing solutions and identifying design gaps that guided our final concept.

Samuele Borgognoni: I completed my bachelor's degree in Mechanical Engineering at Università Politecnica delle Marche (UNIVPM), Ancona, Italy. At UNIVPM I am currently studying Mechanical Engineering, Mechatronics track. Here at VUB I am following an Erasmus+ period of study. For this project, I was mainly involved in the control and software aspects of the robot. My work included defining system requirements, contributing to control logic design, and supporting testing phases to validate autonomous behavior. My favorite part of the project was working on the control strategy, as it required linking theory with practical system behavior.

Nong Deogratius Nevehayi: I completed my Bachelor's in the Department of Mechanical Engineering Option Mehanical Design in the University of Bamenda Cameroon. In this project, I contributed to the system integration, testing, and validation of the robot. I helped ensure that the mechanical, electronic, and software subsystems worked together correctly and supported testing under realistic operating conditions. My favorite part of the project was the final testing phase, where the complete system performance could be evaluated.

Yann Ketis Noubayo Wouassi: I completed my bachelor's degree in civil engineering at ULB. My main role in this project was related to the mechanical design and manufacturing considerations. I worked on CAD modelling and material selection considerations and supported the preparation of parts for prototyping and assembly. My favorite part of the project was seeing the mechanical components take shape from initial designs to physical parts.

Below, the link to the Google Drive folder with the images, the schematics and the material used in the report.

https://drive.google.com/drive/folders/14XviYiA91fEEP4nki3kIp26f_uBiiHEI?usp=sharing