taban/PIR_MadMax
2
4
Fork 0
Code Issues 4 Pull requests Projects Releases Packages Wiki Activity Actions

restructuration pour qu'on puisse travailler ensemble

Browse source
This commit is contained in:
Aleksander Taban 2026-05-14 20:40:08 +02:00
parent 21fa1dfb2a
commit eda53a6535
No known key found for this signature in database
15 changed files with 634 additions and 866 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
Article_Scientifique/IA.tex Normal file
View file

@ -0,0 +1,14 @@
\section*{Statement on AI Usage}
The authors acknowledge the use of generative AI tools during this project, both for the development work and for writing this paper.
AI was used as a helper in several parts of the project. This includes support for understanding and structuring technical ideas, and giving suggestions during the development of different subsystems. It was also used to help with writing, rephrasing, and improving clarity in the report.
However, all final decisions, implementations, and validations were done by the authors. The AI outputs were always checked, corrected when needed, and adapted based on reliable technical sources and our own experiments and understanding of the system.
We consider AI as a useful tool to speed up thinking and writing, but not as a source of final technical truth. Everything related to design choices, analysis, and results was verified and fully controlled by the authors.
The use of AI tools in this work follows the IEEE guidelines for generative AI usage in publications.

14
Article_Scientifique/aim.tex Normal file
View file

@ -0,0 +1,14 @@
\section{Aim and Research Objectives}
This work presents the design and implementation of a motor control system for electric bicycles and cargo transport applications developed within the context of the Manufacture Autonome Décentralisée (MAD) initiative at INSA Toulouse. The main objective is to develop a modular, open-source, and locally manufacturable control architecture adapted to low-cost electric mobility systems.
To achieve this, the project is structured into four main technical contributions.
First, a low-cost motor controller is designed based on a six-step (trapezoidal) commutation strategy. The objective is to eliminate the need for a microcontroller by relying exclusively on discrete MOSFETs and standard electronic components, thereby improving repairability, accessibility, and ease of local manufacturing.
Second, a high-performance controller based on Field-Oriented Control (FOC) is developed using an STM32 microcontroller platform. This implementation leverages and adapts the open-source VESC firmware to ensure compatibility with the selected hardware while enabling advanced motor control capabilities.
Third, the security of the wireless communication interface is investigated, with a focus on Bluetooth Low Energy (BLE) vulnerabilities. A Flipper Zero device is used as a diagnostic tool to evaluate potential attack surfaces and identify weaknesses in the communication layer.
Finally, a dynamic model of the bicycle-cargo system is developed to improve rider experience. The objective is to minimize the perceived additional effort when towing a cargo cart. This is achieved through a PID-based (Proportional-Integral-Derivative) control strategy combined with distance sensing, allowing adaptive assistance based on system dynamics.

17
Article_Scientifique/conclusion.tex Normal file
View file

@ -0,0 +1,17 @@
\section{Conclusion/Summary}
%\begin{figure}[htbp]
%%\centerline{\includegraphics{fig1.png}}
%caption{Example of a figure caption.}
%\label{fig}
%\end{figure}
%Figure Labels: Use 8 point Times New Roman for Figure labels. Use words
%rather than symbols or abbreviations when writing Figure axis labels to
%avoid confusing the reader. As an example, write the quantity
%``Magnetization'', or ``Magnetization, M'', not just ``M''. If including
%units in the label, present them within parentheses. Do not label axes only
%with units. In the example, write ``Magnetization (A/m)'' or ``Magnetization
%\{A[m(1)]\}'', not just ``A/m''. Do not label axes with a ratio of
%quantities and units. For example, write ``Temperature (K)'', not
%``Temperature/K''.

17
Article_Scientifique/discussion.tex Normal file
View file

@ -0,0 +1,17 @@
\section{Discussion}
This project could be seen as an introduction to the VESC project for someone who don't know about it from beforehand, the challenges the new users face during setup, as well as a demand for clear expectations concerning documentation on the subject. The project the MAD is leading should probably not be a fork of the project, as the project is still in development.
As a final note, this proved to be a project which could easily be developed into several different projects in different fields. Some projects could be continued later on as a different PIR subject, other could be proposed to later years in different specialisations like TLS-SEC, ESPE. Our thoughts on the following projects that could be
The fabrication line for electronics is globalised. This is okay in a stable world, but it could be a problem in a world full of instability, be it war, blockages, or tariffs. The idea of opening a specialisation in cooperation with AIME came up as an idea.
For TLS-SEC the subject could be the design for a fitting mechanism to restrict certain privileges to certified personnel that could be used in the C programming language. Later down the line we could also see the possibility to analyse the Bluetooth frames in order to manipulate them in order to change important parameters.
The continuation on the PCB could be a subject fitting an ESPE specialisation.
The proposition of and supply of a VESC system to play with and troubleshoot could be a good rule of thumb, which allows for a quicker start and gives among other things an idea of the budget and the supply line used by a entity in the sector. Proposing a visit could also be one way to familiarise students with the association.
What should be a clear conclusion from our test with the jammer is that a controller based on Bluetooth alone should be avoided when possible and practical. Examples where this could be relevant include electric skateboards, as cables could impose a tripping hazard. There, an encapsulation of an encrypted control frame could be an thought.

116
Article_Scientifique/foc.tex Normal file
View file

@ -0,0 +1,116 @@
\section{STM32-Based Field-Oriented Control Motor Drive}
\label{sec:foc}
This section presents the design and implementation of a high-performance motor controller based on Field-Oriented Control (FOC).
\subsection{Choice of FOC Over Trapezoidal Commutation}
Table~\ref{tab:foc_vs_trap} summarizes the key differences between the two commutation strategies, based on the literature reviewed in Section~\ref{sec:relatedwork}.
\begin{table}[htbp]
\caption{Comparison between FOC and trapezoidal (six-step) commutation}
\label{tab:foc_vs_trap}
\centering
\begin{tabular}{lcc}
\toprule
\textbf{Criterion} & \textbf{FOC} & \textbf{Six-Step} \\
\midrule
Torque ripple (at 500 rpm) & \SI{18.4}{\percent} & \SI{35.7}{\percent} \\
Low-load efficiency & High & Moderate \\
High-speed switching loss & Higher & Lower \\
Position sensor requirement & Encoder (high resolution) & Hall sensors \\
Implementation complexity & High & Low \\
Hardware cost & Higher & Lower \\
Dynamic response & Fast & Standard \\
\bottomrule
\end{tabular}
\end{table}
For our cargo bike application, rider comfort and smooth torque delivery are priorities. FOC was therefore selected for the high-performance controller, while a separate low-tech six-step board (Section~\ref{sec:sixstep}) was developed for repairability.
\subsection{Base Design: Cheap FOCer-2 Project}
The starting point was the open-source \textit{Cheap FOCer-2} project, which provides a complete KiCad design for a VESC-compatible board based on an STM32F405 microcontroller. This design includes:
\begin{itemize}
\item A three-phase MOSFET full-bridge power stage.
\item Gate drivers with built-in dead-time insertion.
\item Shunt resistors for phase current sensing.
\item USB and CAN interfaces.
\item An expansion header for encoder or Hall sensors.
\end{itemize}
The existing KiCad schematic and layout were used as the baseline for our adaptations.
\subsection{Integration of the Rocacher FOC Tile}
Mr. Rocacher provided the Kicad schematic of a ready-to-use FOC tile based on an STM32L476 microcontroller.
The initial idea was to make this tile \textit{pluggable} into our carrier board, similar to an Arduino shield. This would allow :
\begin{itemize}
\item Easy replacement of the computing core without re-soldering.
\item Modular upgrades of the microcontroller.
\item Simplified repair and maintenance.
\end{itemize}
However, the Cheap FOCer-2 project was not designed for such modularity. Its routing is dense and highly optimized for a single, non-removable F405 chip. Adapting it to accept an L476 tile while preserving all critical functions (PWM, current sensing, USB communication) proved challenging.
\subsection{Pin Compatibility Verification: L476 vs F405}
Before starting the PCB modifications, a pin compatibility study was carried out between the STM32L476 used on the Rocacher tile and the STM32F405 originally used in the Cheap FOCer-2 design. The objective was to verify that the main functions required by the VESC firmware could still be used after replacing the original microcontroller.
The verification mainly focused on:
\begin{itemize}
\item Physical pinout compatibility in the LQFP64 package,
\item PWM timer for Alternate functions,
\item USB DP/DM pins (PA11/PA12),
\item Analog inputs for current sensing,
\item UART communication for BLE integration.
\end{itemize}
During this analysis, three main pin conflicts were identified.
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{Figures/CompatibiliteL4F4.pdf}
\caption{Comparison of F405 and L476 pin configurations}
\end{figure}
The first conflict concerned the SPI\_MISO signal on pin PA6. In the original STM32F405 design, this pin is used for SPI communication related to current sensing. On the STM32L476 tile, the same pin is associated with a DAC output, creating a functional conflict. To solve this issue, the SPI communication line was remapped to PA5 on the L476, which offers a compatible alternate function.
The second issue concerned the EN\_GATE signal. In the original design, this signal was connected to PB5 on the STM32F405. However, this pin is not accessible on the L476 tile. The signal was therefore moved to PC5, configured as a standard GPIO output.
Finally, Hall sensor C was originally connected to PC8 (TIM8) on the STM32F405. Since this pin is not available on the tile connector, the Hall sensor input was reassigned to PB3 using the TIM2\_CH2 alternate function, which preserves the input capture capability required for Hall sensor decoding.
All other important functions remained compatible between the two microcontrollers, including PWM generation, complementary PWM outputs, encoder inputs, UART, USB, and CAN communication. Some differences between the ADC peripherals of the STM32F405 and STM32L476 still remain and will require firmware adaptations in future work.
\subsection{Schematic Design and KiCad Implementation}
The original Cheap FOCer-2 schematic was modified in KiCad in order to replace the integrated STM32F405 microcontroller with connectors for the Rocacher STM32L476 tile. The objective was to make the control part more modular and easier to replace without modifying the power stage of the board.
The main modifications performed on the schematic were:
\begin{itemize}
\item Removal of the STM32F405 and its associated passive components.
\item Addition of two 20-pin headers for the L476 tile connection.
\item Re-routing of PWM, ADC, USB, and communication signals toward the headers.
\end{itemize}
Special attention was given to the routing of critical control signals, especially the PWM outputs used for motor commutation and the analogue signals used for current sensing.
After the modifications, the schematic was verified using the KiCad Electrical Rule Check (ERC). No electrical errors were detected during this verification step, which validated the consistency of the schematic before starting the PCB routing phase.
\subsection{Routing Challenges and Current Status}
After validating the schematic, the PCB routing phase was started in KiCad. The original Cheap FOCer-2 board uses a very compact layout with dense routing around the STM32F405 microcontroller and the power stage. Integrating connectors for a removable STM32L476 tile introduced several additional routing constraints.
One of the main difficulties was maintaining proper signal routing while keeping enough space for the tile connectors and preserving the integrity of the control signals. Particular attention had to be given to the PWM signals, current sensing traces, and power connections.
Several issues were encountered during the routing process:
\begin{itemize}
\item Some connector footprints associated with the tile did not appear correctly after importing the schematic into the PCB layout.
\item The routing of high-current paths, especially the battery and motor phase connections, become more complex due to the additional connectors and required extra vias.
\item Some Decoupling capacitors had to be repositioned, which could potentially affect switching noise and power
supply stability.
\end{itemize}
At the current stage of the project, the schematic has been validated and the PCB layout is still under development. Once the routing is completed, the board will be manufactured and tested using the VESC firmware adapted for the STM32L476 tile.

14
Article_Scientifique/futur.tex Normal file
View file

@ -0,0 +1,14 @@
\section{Perspectives and Future Work}
Based on the results obtained and the limitations identified during this project, several directions for future work are proposed.
\subsection{Hardware Completion and Testing}
The VESC-based FOC PCB requires routing completion and prototype manufacturing. Once fabricated, the board must be tested under real operating conditions (varying loads, road profiles, and battery voltage).
\subsection{Control Strategy Enhancement}
Future work includes the implementation and comparison of the PI controller and the cascaded control structure on the cargo cart, in order to evaluate their performance under different conditions (load variations, road profiles, etc.).
In addition, the lack of direct velocity measurements limited the ability to fully characterise the system dynamics. Adding a speed measurement or improving state estimation would allow a more complete analysis of the model.

11
Article_Scientifique/gap.tex Normal file
View file

@ -0,0 +1,11 @@
\section{Research Gap}
The literature presented in the previous sections shows that significant work has been carried out on BLDC motor control strategies, especially for torque ripple reduction, efficiency improvement, and dynamic performance optimisation. Open-source projects such as VESC also provide high-performance and flexible solutions for electric mobility systems.
However, most existing works mainly focus on control performance and do not consider low-tech constraints such as local manufacturing, hardware repairability, or component accessibility. In many existing controllers, the hardware design remains difficult to reproduce or repair without specialized equipment or advanced electronic knowledge.
In addition, the integration of wireless communication introduces new security concerns. While BLE connectivity simplifies configuration and monitoring, unauthorized access to controller parameters could create safety risks, especially for electric cargo bikes operating in public spaces.
To the best of our knowledge, there is currently no open-source motor controller that simultaneously addresses high-performance control, local manufacturability, repairability, and BLE security for decentralized cargo bike applications.
This project therefore aims to explore a modular and repair-oriented motor controller architecture compatible with VESC while remaining adapted to the constraints of the Manufacture Autonome Décentralisée (MAD).

10
Article_Scientifique/intro.tex Normal file
View file

@ -0,0 +1,10 @@
\section{Introduction}
The fast urbanization of global logistics has positioned electric cargo bikes as a primary solution. At the heart of these vehicles is the motor controller. Current research and industry standards primarily focus on two methods of commutation for the controller: Trapezoidal commutation and Field Oriented Control (FOC).
As motor controllers become smarter, they increasingly incorporate wireless connectivity for tuning and diagnostics. Current research highlights that while Bluetooth Low Energy (BLE) and mobile app integration improve user experience, they often introduce vulnerabilities. Open-source projects, in particular, must balance ease of access for community developers with the need to secure the vehicle.
We also argue the need for general public's safety when it comes to these bikes, as it could be a danger to the traffic. This is especially true when it comes to vehicles carrying a substantial load. This needs to be considered by the MAD, where their responsibility and control begins and ends. Should there be a difference between the firmware loaded on a product from the MAD than what is publicly available?

135
Article_Scientifique/logiciel.tex Normal file
View file

@ -0,0 +1,135 @@
\section{Software and Connectivity}
\subsection{BLE Compatibility With the VESC}
\subsubsection{First Experiment}
VESC-controllers are not necessarily equipped with Bluetooth-modules by default. Often, it is necessary to add a BLE-module. A standard HC-05 bluetooth-module compatible with Arduino is a great way to send and receive bluetooth-packets from a host, e.g. a mobile phone, via a bridge translating the bluetooth packets to the UART protocol. This could be demonstrated using a ESP8622's standard library with said module, by letting us send characters from one device to another.
\subsubsection{HC-05 and the VESC}
By flashing the VESC firmware on a discovery-card and connecting the HC-05 module to the PB10 and PB11-pins, which are the Rx and Tx-pins for the STM32F4xx chip, we discovered that the setup for the bluetooth module was not available in the VESC tool. The inherent BLE capabilities is an important limitation to consider when designing a VESC system. We learned therefore that the HC-05 is not originally adapted for BLE. The need for a bridge also adds on complexity and cost, in the form of extra components and another device to maintain the code of. For the future, choosing a bluetooth module supporting BLE will be the easiest solution. Preferably a module fitting the communication connector on the cheap FOCer project\cite{b1} could facilitate the relevancy of the PCB project with a microcontroller.
\subsubsection{BLE Vulnerability}
Bluetooth could be a vulnerability to a VESC if it is to be used as a controller in real-time, as the controller could be jammed. Our test with the Flipper Zero shows the disfunctionnality of Bluetooth with different use cases. We experienced with the jamming of a bluetooth speaker that the music completely stopped. It could also be investigated how the connection to the VESC could be modified using the VESC tool. We will touch more on the accessibility of the code within the VESC tool sooner.
\subsection{Setup of BT-connection with VESC}
The setup consisted of a main PC/controller, HC-05 Bluetooth module, ESP8266 µ-controller, STM32 Discovery µ-controller, PC with VESC-tool as well as a \textit{FlipperZero}. Our plan of action consisted of flashing the Discovery-card with code for then to read this code via UART to the ESP8266 which was connected to the BT-module. The BT-module would then send packets to the PC. This PC would then act as read/write to read the code having been flashed on the Discovery. Full-band jamming would then be achieved by the implementation of the \textit{FlipperZero} disrupting any transfer of code from PC1 to PC2.
\newpage
\textbf{Setup of BT-connection with VESC}
\begin{tikzpicture}
% PC_ESP
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (PC1) at (2.5,-0.5)
{
\includegraphics[width=1.5cm]{Figures/PC_STOCK.jpg}
};
\node[above=2mm of PC1] {PC1 ready to r/w};
% HC05
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (HC05) at (-0.3,-2.6)
{
\includegraphics[width=1.5cm]{Figures/HC-05_Bluetooth_Module.jpg}
};
\node[above=1mm of HC05] {HC-05};
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (ESP) at (-0.3,-5)
{
\includegraphics[width=1.5cm]{Figures/ESP8266.png}
};
\node[below=1mm of ESP] {ESP8266};
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (PC2) at (-3.5,-2.6)
{
\includegraphics[width=1.5cm]{Figures/PC_STOCK.jpg}
};
\node[above=1mm of PC2] {PC2 with VESC-tool};
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (Disc) at (-3.5,-5)
{
\includegraphics[width=1.5cm]{Figures/STM32_Discovery.jpg}
};
\node[below=1mm of Disc] {STM32 Discovery};
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (F0) at (-1.5,0)
{
\includegraphics[width=1.5cm]{Figures/F0.jpeg}
};
\node[above=1mm of F0] {FlipperZero};
% Arrow between boxes
%\draw[decorate, decoration={snake}] (box1) -- (box2);
\draw[->, thick] (HC05) -- (ESP);
\draw[<->, decorate, decoration={snake}] (PC1) -- node[below right]{BT-connectivity} (HC05);
\draw[->] (HC05) -- node[left]{UART} (ESP);\\
\draw[<->] (HC05) -- node[right]{Tx $\leftrightarrow$ Rx} (ESP);
\draw[<->] (ESP) -- node[below]{UART} (Disc);
\draw[<->] (ESP) -- node[above]{Tx $\leftrightarrow$ Rx} (Disc);
\draw[<->] (Disc) -- node[left]{USB} (PC2);
\draw[->, decorate, decoration={snake}] (F0) -- node[above=3mm]{Jamming} (1.25,-1.5);
\
\end{tikzpicture}
\subsection{Code integrity}
\subsubsection{Context}
As the project is open source, and the code is freely accessible, there should be no reason to hide the code. It could however be reasonable to protect the code from changes which could hurt other people. Changing following parameters should at least come with a disclaimer and clearly state the dangers possible by proceeding with said changes. We have in mind the maximum speed permitted and the power available to the motors.
\subsubsection{LispBM extraction}
We caught word that the lisp code for the VESC used by Maillon mobility was easy to extract. By building an older firmware with the Maillon mobility software, we observed this by going to the lispBM tab and clicking read. It's up to the MAD if they would like to reinforce this mechanism. A modification on a parameter and then clicking upload allowed us to easily change the speed limit. This could bring up a public danger. This raises questions on the use of the MAD's equipment which is in a traffic friendly manner.
\subsubsection{LispBM Code}
When we flashed newer firmware from the project made by Benjamin Vedder\cite{b1}, we also observed some difficulties in uploading the lispBM script taken from the one on firmware version 6.06. This could indicate that there needs to be further maintenance of the code in order to get the software up to speed. This needs to be documented better for someone to continue the project. This could be a good investment for the MAD as well in the context of training for the people working on the motor control part of the e-bike.
This documentation could be as simple as referencing the relevant parts of the lispBM documentation \cite{b2}
\subsubsection{Proposed Solution}
This risk could be patched by developing a VESC application for the VESC controller or using a binary. This is a solution which is less open source, but which is make unlawful use of the material harder. The application could be created using C and use an algorithm known by the MAD in order to secure the access to someone to change the parameters only if they are the MAD certified personnel. This encryption would preferably be reduced to the most essential settings in order to align with what our impression of the philosophy of the MAD would be.
\subsection{VESC Compiling}
As mentioned, we have been able to compile the VESC tool and the VESC firmware. This firmware has been put onto an STM32F4xx Discovery card. This card uses the same chip as the aforementioned ``Cheap FOCer'' project. The thought was that using something with the same chip would facilitate the switch from the discovery card to a PCB with the same target.
However, this choice posed several obstacles for our progress on the topic of cybersecurity. We will nonetheless summarise what we have learned for you and propose some additional work for the future. The challenges we encountered were the following: The lack of bluetooth capabilities. We did not have a module with BLE either. We had access to a HC-05 module, but that only allows for a normal bluetooth protocol and would require further work on a bridge to UART by using an esp8622 that we had as well. We propose that the next group has access to a VESC controller from the beginning, as well as a motor we could control. This could be in cooperation with the MAD, as the MAD could propose some models they're interested in.
We also found that the information on the VESC is scattered around the internet. The resources is also sometimes based on a Debian-based Linux system which adds more work for someone using another distribution of Linux. This could hinder the implementation facility for new users. We struggled particularly with the Qt packages for positioning and game pad. We would therefore recommend the use of a Debian-based Linux system for the computer working with the VESC for the MAD associates.

26
Article_Scientifique/lowtech.tex Normal file
View file

@ -0,0 +1,26 @@
\section{Hardware-Based Six-Step Commutation Controller}
\label{sec:sixstep}
The basis of this section is the replacement of components who are highly complex, technical and/or are dependant on global supply chains to manufacture. An additional goal is, as for the preceding section, to make a repairable, reliable and manufacturable circuit, this time using these more basic components, based on open-source principles. The controller needs to be performant enough to drive one of the two electric motors used on the LaMAD (La Manufacture Autonome Décentralisée) bicycle cargo trailer. %ref moteur qu'utilise lamad
\subsection{AIME and other SC-facilities capabilities}
The most challenging part of this section is the replacement of semiconductor parts.
\subsection{Replacing an IC}
Replacing the IC of a motor controller requires using traditional logic gates. This approach is
\subsection{Power components}
\subsection{Clock}
To calculate the maximum rotation speed, we take the worst conditions of the motor used in the MAD cargo bike; 50 km/h\footnote{http://www.mxusebikekit.com/pro\_info.asp?Pid=25} with a 50 cm wheel\footnote{http://www.mxusebikekit.com/pro\_info.asp?Pid=25}\footnote{https://veloma.org/2022/10/05/la-charrette-version-montagne-ou-comment-transporter-250kg-a-velo-par-monts-et-par-vaux/} (not measured, assumed from the images and the motor specifications).
\begin{equation*}
\omega=\frac{v}{\pi d}=\frac{50\cdot 10^3m/h\cdot\frac{1}{60}h/min}{3,14\cdot 50\cdot 10^{-2}m}=530\ rpm
\end{equation*}
The motor being a three-phase brushless motor, the switching speed needs to be

886
Article_Scientifique/main.tex
View file

@ -97,873 +97,27 @@ bikes while significantly improving repairability.
VESC, Brushless DC motor, Field Oriented Control, Trapezoidal commutation, Low-Tech, PID-Control.
\end{IEEEkeywords}
\section{Introduction}
The fast urbanization of global logistics has positioned electric cargo bikes as a primary solution. At the heart of
these vehicles is the motor controller. Current research and industry standards primarily focus on two methods of
commutation for the controller: Trapezoidal commutation and Field Oriented Control (FOC).
As motor controllers become smarter, they increasingly incorporate wireless connectivity for tuning and diagnostics.
Current research highlights that while Bluetooth Low Energy (BLE) and mobile app integration improve user experience,
they often introduce vulnerabilities. Open-source projects, in particular, must balance ease of access for community
developers with the need to secure the vehicle.
We also argue the need for general public's safety when it comes to these bikes, as it could be a danger to the traffic.
This is especially true when it comes to vehicles carrying a substantial load. This needs to be considered by the MAD,
where their responsibility and control begins and ends. Should there be a difference between the firmware loaded on a
product from the MAD than what is publicly available?
% ********************************************* RELATED WORK ***********************************************************
\section{Related Work}
\label{sec:relatedwork}
\subsection{Modeling of BLDC Motor}
The electromechanical model of a BLDC (Brushless DC) motor is foundational for understanding its behaviour under different control
schemes. BLDC motors are categorized by their back-electromotive force (back-EMF) waveform: trapezoidal or sinusoidal.
This distinction is crucial, as the trapezoidal shape inherently leads to torque ripple when the supplied phase currents
are not perfectly aligned, directly influencing the choice and effectiveness of the control strategy
\cite{patil_analysis_2025}. For a BLDC motor with trapezoidal back-EMF, the electromagnetic torque is given by:
\[
T_e = \frac{e_a i_a + e_b i_b + e_c i_c}{\omega_m}
\]
where \( e_x \) is the back-EMF and \( i_x \) is the phase current \cite{li_quantitative_2019}. The classical d-q
reference frame model, ideal for sinusoidal machines, is less suitable for trapezoidal BLDC motors because it assumes
sinusoidal flux distribution. Phase-variable modelling in the natural (abc) frame is therefore more appropriate, as it
directly accounts for the non-sinusoidal, trapezoidal nature of the back-EMF and the associated harmonics
\cite{mohammd_taher_new_2021}.
\subsection{Trapezoidal Commutation (Six-Step Control) for BLDC Motors}
% Trapezoidal commutation, or six-step control, uses Hall-effect sensors to synchronize phase current switching every
% 120 electrical degrees.
Trapezoidal commutation, or Six-Step control, uses bipolar conduction, with two motor phases conducting at any time and
current commutation occurring every 120 electrical degrees \cite{gieras_modern_2023}. As commutation depends on rotor
position, Six-Step control requires either position sensors (e.g. Hall sensors, encoders, or resolvers) or sensor-less
estimation based on back-EMF detection or observers \cite{gieras_modern_2023, gasc_conception_2004}. This method is
renowned for its simplicity of implementation and low hardware cost \cite{bhatiya_bldc_2024}. It enables effective
torque control but introduces significant torque ripple during commutation events, especially under high load
\cite{jomsa-nga_torque_2024}. This ripple generates noise, increases mechanical stress, and reduces overall efficiency
\cite{mohammd_taher_new_2021}. Although PWM techniques can mitigate this ripple, they do not completely eliminate it
\cite{li_quantitative_2019}.
\subsection{Field-Oriented Control (FOC) for BLDC Motors}
FOC is a vector control strategy that decouples the stator flux and torque components. It transforms three-phase
currents into orthogonal \( I_d \) and \( I_q \) components, enabling precise torque control and significant ripple
reduction \cite{jomsa-nga_torque_2024}. FOC is particularly effective for BLDC motors with sinusoidal back-EMF but can
also be applied to trapezoidal back-EMF motors, albeit with less impressive ripple suppression results
\cite{li_quantitative_2019}. It requires greater computational power and more precise position sensors (e.g. encoders).
Comparative analysis shows that FOC yields a more stable stator current profile and significantly reduces torque
variations compared to trapezoidal control \cite{patil_analysis_2025}.
\subsection{Comparative Analysis: FOC vs. Trapezoidal for Light Electric Vehicles}
\subsubsection{Torque Ripple and User Comfort}
Firstly, torque ripple can be reduced for both control methods by selecting appropriate motor parameters, such as the
number of stator slots and rotor poles \cite{gasc_conception_2004}.
FOC substantially reduces torque ripple compared to Six-Step control, directly enhancing ride comfort and minimizing
vibrations. Experimental results show a torque ripple of \SI{18.38}{\percent} for FOC versus \SI{35.67}{\percent} for
Six-Step control at 500~rpm \cite{jomsa-nga_torque_2024}.
Commutation torque ripple (CTR), prominent in Six-Step control, can be specifically targeted and mitigated using
advanced control techniques like Model Predictive Control (MPC) while retaining the fundamental simplicity of
trapezoidal commutation \cite{mohammd_taher_new_2021}.
\subsubsection{Energy Efficiency}
FOC optimizes torque per ampere (MTPA), improving efficiency at low loads. Six-Step control exhibits lower switching
losses at high speeds \cite{li_quantitative_2019}.
\subsubsection{Complexity, Cost, and Low-Tech Suitability}
Six-Step control is inherently simpler, cheaper, and more robust, making it a prime candidate for low-tech applications.
Research focused on reducing propulsion system costs proposes simplified hardware topologies, such as 4-switch inverters
(instead of 6) coupled with direct current control strategies, maintaining acceptable performance while significantly
lowering hardware costs \cite{lee_advanced_2001}. FOC, while superior in performance, is more complex to implement and
carries higher hardware costs (sensors, processing power).
\subsubsection{Dynamic Response}
FOC provides faster response times and better load disturbance rejection \cite{jomsa-nga_torque_2024}.
\section{Research Gap}
The literature presented in the previous sections shows that significant work has been carried out on BLDC
motor control strategies, especially for torque ripple reduction, efficiency improvement, and dynamic performance
optimisation. Open-source projects such as VESC also provide high-performance and flexible solutions for electric
mobility systems.
However, most existing works mainly focus on control performance and do not consider low-tech constraints such as
local manufacturing, hardware repairability, or component accessibility. In many existing controllers, the hardware
design remains difficult to reproduce or repair without specialized equipment or advanced electronic knowledge.
In addition, the integration of wireless communication introduces new security concerns. While BLE connectivity
simplifies configuration and monitoring, unauthorized access to controller parameters could create safety risks,
especially for electric cargo bikes operating in public spaces.
To the best of our knowledge, there is currently no open-source motor controller that simultaneously addresses
high-performance control, local manufacturability, repairability, and BLE security for decentralized cargo bike
applications.
This project therefore aims to explore a modular and repair-oriented motor controller architecture compatible
with VESC while remaining adapted to the constraints of the Manufacture Autonome Décentralisée (MAD).
\section{Aim and Research Objectives}
This work presents the design and implementation of a motor control system for electric bicycles and cargo transport
applications developed within the context of the Manufacture Autonome Décentralisée (MAD) initiative at INSA Toulouse.
The main objective is to develop a modular, open-source, and locally manufacturable control architecture adapted to
low-cost electric mobility systems.
To achieve this, the project is structured into four main technical contributions.
First, a low-cost motor controller is designed based on a six-step (trapezoidal) commutation strategy. The objective is
to eliminate the need for a microcontroller by relying exclusively on discrete MOSFETs and standard electronic
components, thereby improving repairability, accessibility, and ease of local manufacturing.
Second, a high-performance controller based on Field-Oriented Control (FOC) is developed using an STM32 microcontroller
platform. This implementation leverages and adapts the open-source VESC firmware to ensure compatibility with the
selected hardware while enabling advanced motor control capabilities.
Third, the security of the wireless communication interface is investigated, with a focus on Bluetooth Low Energy (BLE)
vulnerabilities. A Flipper Zero device is used as a diagnostic tool to evaluate potential attack surfaces and identify
weaknesses in the communication layer.
Finally, a dynamic model of the bicycle-cargo system is developed to improve rider experience.
The objective is to minimize the perceived additional effort when towing a cargo cart. This is achieved through a
PID-based (Proportional-Integral-Derivative) control strategy combined with distance sensing, allowing adaptive
assistance based on system dynamics.
% ************************************** FIELD ORIENTED CONTROL *****************************************************
\section{STM32-Based Field-Oriented Control Motor Drive}
\label{sec:foc}
This section presents the design and implementation of a high-performance
motor controller based on Field-Oriented Control (FOC).
\subsection{Choice of FOC Over Trapezoidal Commutation}
Table~\ref{tab:foc_vs_trap} summarizes the key differences between the
two commutation strategies, based on the literature reviewed in
Section~\ref{sec:relatedwork}.
\begin{table}[htbp]
\caption{Comparison between FOC and trapezoidal (six-step) commutation}
\label{tab:foc_vs_trap}
\centering
\begin{tabular}{lcc}
\toprule
\textbf{Criterion} & \textbf{FOC} & \textbf{Six-Step} \\
\midrule
Torque ripple (at 500 rpm) & \SI{18.4}{\percent} & \SI{35.7}{\percent} \\
Low-load efficiency & High & Moderate \\
High-speed switching loss & Higher & Lower \\
Position sensor requirement & Encoder (high resolution) & Hall sensors \\
Implementation complexity & High & Low \\
Hardware cost & Higher & Lower \\
Dynamic response & Fast & Standard \\
\bottomrule
\end{tabular}
\end{table}
For our cargo bike application, rider comfort and smooth torque delivery
are priorities. FOC was therefore selected for the high-performance
controller, while a separate low-tech six-step board (Section~\ref{sec:sixstep})
was developed for repairability.
\subsection{Base Design: Cheap FOCer-2 Project}
The starting point was the open-source \textit{Cheap FOCer-2} project,
which provides a complete KiCad design for a VESC-compatible board based
on an STM32F405 microcontroller. This design includes:
\begin{itemize}
\item A three-phase MOSFET full-bridge power stage.
\item Gate drivers with built-in dead-time insertion.
\item Shunt resistors for phase current sensing.
\item USB and CAN interfaces.
\item An expansion header for encoder or Hall sensors.
\end{itemize}
The existing KiCad schematic and layout were used as the baseline for
our adaptations.
\subsection{Integration of the Rocacher FOC Tile}
Mr. Rocacher provided the Kicad schematic of a ready-to-use FOC tile based on an STM32L476
microcontroller.
The initial idea was to make this tile \textit{pluggable} into our
carrier board, similar to an Arduino shield. This would allow :
\begin{itemize}
\item Easy replacement of the computing core without re-soldering.
\item Modular upgrades of the microcontroller.
\item Simplified repair and maintenance.
\end{itemize}
However, the Cheap FOCer-2 project was not designed for such modularity.
Its routing is dense and highly optimized for a single, non-removable
F405 chip. Adapting it to accept an L476 tile while preserving all
critical functions (PWM, current sensing, USB communication) proved
challenging.
\subsection{Pin Compatibility Verification: L476 vs F405}
Before starting the PCB modifications, a pin compatibility study was carried out between the STM32L476
used on the Rocacher tile and the STM32F405 originally used in the Cheap FOCer-2 design. The objective
was to verify that the main functions required by the VESC firmware could still be used after replacing
the original microcontroller.
The verification mainly focused on:
\begin{itemize}
\item Physical pinout compatibility in the LQFP64 package,
\item PWM timer for Alternate functions,
\item USB DP/DM pins (PA11/PA12),
\item Analog inputs for current sensing,
\item UART communication for BLE integration.
\end{itemize}
During this analysis, three main pin conflicts were identified.
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{Figures/CompatibiliteL4F4.pdf}
\caption{Comparison of F405 and L476 pin configurations}
\end{figure}
The first conflict concerned the SPI\_MISO signal on pin PA6. In the original STM32F405 design, this pin is
used for SPI communication related to current sensing. On the STM32L476 tile, the same pin is associated with
a DAC output, creating a functional conflict. To solve this issue, the SPI communication line was remapped
to PA5 on the L476, which offers a compatible alternate function.
The second issue concerned the EN\_GATE signal. In the original design, this signal was connected to PB5 on
the STM32F405. However, this pin is not accessible on the L476 tile. The signal was therefore moved to PC5,
configured as a standard GPIO output.
Finally, Hall sensor C was originally connected to PC8 (TIM8) on the STM32F405. Since this pin is not available
on the tile connector, the Hall sensor input was reassigned to PB3 using the TIM2\_CH2 alternate function,
which preserves the input capture capability required for Hall sensor decoding.
All other important functions remained compatible between the two microcontrollers, including PWM generation,
complementary PWM outputs, encoder inputs, UART, USB, and CAN communication. Some differences between the ADC
peripherals of the STM32F405 and STM32L476 still remain and will require firmware adaptations in future work.
\subsection{Schematic Design and KiCad Implementation}
The original Cheap FOCer-2 schematic was modified in KiCad in order to replace the integrated STM32F405
microcontroller with connectors for the Rocacher STM32L476 tile. The objective was to make the control part more
modular and easier to replace without modifying the power stage of the board.
The main modifications performed on the schematic were:
\begin{itemize}
\item Removal of the STM32F405 and its associated passive components.
\item Addition of two 20-pin headers for the L476 tile connection.
\item Re-routing of PWM, ADC, USB, and communication signals toward the headers.
\end{itemize}
Special attention was given to the routing of critical control signals, especially the PWM outputs used for
motor commutation and the analogue signals used for current sensing.
After the modifications, the schematic was verified using the KiCad Electrical Rule Check (ERC). No electrical
errors were detected during this verification step, which validated the consistency of the schematic before
starting the PCB routing phase.
\subsection{Routing Challenges and Current Status}
After validating the schematic, the PCB routing phase was started in KiCad. The original Cheap FOCer-2 board
uses a very compact layout with dense routing around the STM32F405 microcontroller and the power stage.
Integrating connectors for a removable STM32L476 tile introduced several additional routing constraints.
One of the main difficulties was maintaining proper signal routing while keeping enough space for the tile
connectors and preserving the integrity of the control signals. Particular attention had to be given to the PWM
signals, current sensing traces, and power connections.
Several issues were encountered during the routing process:
\begin{itemize}
\item Some connector footprints associated with the tile did not appear correctly after importing the schematic
into the PCB layout.
\item The routing of high-current paths, especially the battery and motor phase connections, become more complex
due to the additional connectors and required extra vias.
\item Some Decoupling capacitors had to be repositioned, which could potentially affect switching noise and power
supply stability.
\end{itemize}
At the current stage of the project, the schematic has been validated and the PCB layout is still under development.
Once the routing is completed, the board will be manufactured and tested using the VESC firmware adapted for the
STM32L476 tile.
% ************************************** LOW TECH SIX STEP CONTROL *****************************************************
\section{Hardware-Based Six-Step Commutation Controller}
\label{sec:sixstep}
The basis of this section is the replacement of components who are highly complex, technical and/or are dependant on global supply chains to manufacture. An additional goal is, as for the preceding section, to make a repairable, reliable and manufacturable circuit, this time using these more basic components, based on open-source principles. The controller needs to be performant enough to drive one of the two electric motors used on the LaMAD (La Manufacture Autonome Décentralisée) bicycle cargo trailer. %ref moteur qu'utilise lamad
\subsection{AIME and other SC-facilities capabilities}
The most challenging part of this section is the replacement of semiconductor parts.
\subsection{Replacing an IC}
Replacing the IC of a motor controller requires using traditional logic gates. This approach is
\subsection{Power components}
\subsection{Clock}
To calculate the maximum rotation speed, we take the worst conditions of the motor used in the MAD cargo bike; 50 km/h\footnote{http://www.mxusebikekit.com/pro\_info.asp?Pid=25} with a 50 cm wheel\footnote{http://www.mxusebikekit.com/pro\_info.asp?Pid=25}\footnote{https://veloma.org/2022/10/05/la-charrette-version-montagne-ou-comment-transporter-250kg-a-velo-par-monts-et-par-vaux/} (not measured, assumed from the images and the motor specifications).
\begin{equation*}
\omega=\frac{v}{\pi d}=\frac{50\cdot 10^3m/h\cdot\frac{1}{60}h/min}{3,14\cdot 50\cdot 10^{-2}m}=530\ rpm
\end{equation*}
The motor being a three-phase brushless motor, the switching speed needs to be
% ************************************** SOFTWARE AND CONNECTIVITY *****************************************************
\section{Software and Connectivity}
\subsection{BLE Compatibility With the VESC}
\subsubsection{First Experiment}
VESC-controllers are not necessarily equipped with Bluetooth-modules by default. Often, it is necessary to add a
BLE-module. A standard HC-05 bluetooth-module compatible with Arduino is a great way to send and receive
bluetooth-packets from a host, e.g. a mobile phone, via a bridge translating the bluetooth packets to the UART protocol.
This could be demonstrated using a ESP8622's standard library with said module, by letting us send characters from one
device to another.
\subsubsection{HC-05 and the VESC}
By flashing the VESC firmware on a discovery-card and connecting the HC-05 module to the PB10 and PB11-pins, which are
the Rx and Tx-pins for the STM32F4xx chip, we discovered that the setup for the bluetooth module was not available in
the VESC tool. The inherent BLE capabilities is an important limitation to consider when designing a VESC system.
We learned therefore that the HC-05 is not originally adapted for BLE. The need for a bridge also adds on complexity
and cost, in the form of extra components and another device to maintain the code of. For the future, choosing a
bluetooth module supporting BLE will be the easiest solution. Preferably a module fitting the communication connector on
the cheap FOCer project\cite{b1} could facilitate the relevancy of the PCB project with a microcontroller.
\subsubsection{BLE Vulnerability}
Bluetooth could be a vulnerability to a VESC if it is to be used as a controller in real-time, as the controller could
be jammed. Our test with the Flipper Zero shows the disfunctionnality of Bluetooth with different use cases.
We experienced with the jamming of a bluetooth speaker that the music completely stopped. It could also be investigated
how the connection to the VESC could be modified using the VESC tool. We will touch more on the accessibility of the
code within the VESC tool sooner.
\subsection{Setup of BT-connection with VESC}
The setup consisted of a main PC/controller, HC-05 Bluetooth module, ESP8266 µ-controller, STM32 Discovery µ-controller, PC with VESC-tool as well as a \textit{FlipperZero}. Our plan of action consisted of flashing the Discovery-card with code for then to read this code via UART to the ESP8266 which was connected to the BT-module. The BT-module would then send packets to the PC. This PC would then act as read/write to read the code having been flashed on the Discovery. Full-band jamming would then be achieved by the implementation of the \textit{FlipperZero} disrupting any transfer of code from PC1 to PC2.
\newpage
\textbf{Setup of BT-connection with VESC}
\begin{tikzpicture}
% PC_ESP
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (PC1) at (2.5,-0.5)
{
\includegraphics[width=1.5cm]{Figures/PC_STOCK.jpg}
};
\node[above=2mm of PC1] {PC1 ready to r/w};
% HC05
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (HC05) at (-0.3,-2.6)
{
\includegraphics[width=1.5cm]{Figures/HC-05_Bluetooth_Module.jpg}
};
\node[above=1mm of HC05] {HC-05};
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (ESP) at (-0.3,-5)
{
\includegraphics[width=1.5cm]{Figures/ESP8266.png}
};
\node[below=1mm of ESP] {ESP8266};
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (PC2) at (-3.5,-2.6)
{
\includegraphics[width=1.5cm]{Figures/PC_STOCK.jpg}
};
\node[above=1mm of PC2] {PC2 with VESC-tool};
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (Disc) at (-3.5,-5)
{
\includegraphics[width=1.5cm]{Figures/STM32_Discovery.jpg}
};
\node[below=1mm of Disc] {STM32 Discovery};
\node[
draw,
rectangle,
minimum width=1.5cm,
minimum height=1.5cm
] (F0) at (-1.5,0)
{
\includegraphics[width=1.5cm]{Figures/F0.jpeg}
};
\node[above=1mm of F0] {FlipperZero};
% Arrow between boxes
%\draw[decorate, decoration={snake}] (box1) -- (box2);
\draw[->, thick] (HC05) -- (ESP);
\draw[<->, decorate, decoration={snake}] (PC1) -- node[below right]{BT-connectivity} (HC05);
\draw[->] (HC05) -- node[left]{UART} (ESP);\\
\draw[<->] (HC05) -- node[right]{Tx $\leftrightarrow$ Rx} (ESP);
\draw[<->] (ESP) -- node[below]{UART} (Disc);
\draw[<->] (ESP) -- node[above]{Tx $\leftrightarrow$ Rx} (Disc);
\draw[<->] (Disc) -- node[left]{USB} (PC2);
\draw[->, decorate, decoration={snake}] (F0) -- node[above=3mm]{Jamming} (1.25,-1.5);
\
\end{tikzpicture}
\subsection{Code integrity}
\subsubsection{Context}
As the project is open source, and the code is freely accessible, there should be no reason to hide the code. It could
however be reasonable to protect the code from changes which could hurt other people. Changing following parameters
should at least come with a disclaimer and clearly state the dangers possible by proceeding with said changes. We
have in mind the maximum speed permitted and the power available to the motors.
\subsubsection{LispBM extraction}
We caught word that the lisp code for the VESC used by Maillon mobility was easy to extract. By building an older
firmware with the Maillon mobility software, we observed this by going to the lispBM tab and clicking read. It's up to
the MAD if they would like to reinforce this mechanism. A modification on a parameter and then clicking upload allowed
us to easily change the speed limit. This could bring up a public danger. This raises questions on the use of the MAD's
equipment which is in a traffic friendly manner.
\subsubsection{LispBM Code}
When we flashed newer firmware from the project made by Benjamin Vedder\cite{b1}, we also observed some difficulties in
uploading the lispBM script taken from the one on firmware version 6.06. This could indicate that there needs to be
further maintenance of the code in order to get the software up to speed. This needs to be documented better for someone
to continue the project. This could be a good investment for the MAD as well in the context of training for the people
working on the motor control part of the e-bike.
This documentation could be as simple as referencing the relevant parts of the lispBM documentation \cite{b2}
\subsubsection{Proposed Solution}
This risk could be patched by developing a VESC application for the VESC controller or using a binary. This is a
solution which is less open source, but which is make unlawful use of the material harder. The application could be
created using C and use an algorithm known by the MAD in order to secure the access to someone to change the parameters
only if they are the MAD certified personnel. This encryption would preferably be reduced to the most essential settings
in order to align with what our impression of the philosophy of the MAD would be.
\subsection{VESC Compiling}
As mentioned, we have been able to compile the VESC tool and the VESC firmware. This firmware has been put onto an
STM32F4xx Discovery card. This card uses the same chip as the aforementioned ``Cheap FOCer'' project. The thought was
that using something with the same chip would facilitate the switch from the discovery card to a PCB with the same target.
However, this choice posed several obstacles for our progress on the topic of cybersecurity. We will nonetheless
summarise what we have learned for you and propose some additional work for the future. The challenges we encountered
were the following: The lack of bluetooth capabilities. We did not have a module with BLE either. We had access to a
HC-05 module, but that only allows for a normal bluetooth protocol and would require further work on a bridge to UART
by using an esp8622 that we had as well. We propose that the next group has access to a VESC controller from the
beginning, as well as a motor we could control. This could be in cooperation with the MAD, as the MAD could propose
some models they're interested in.
We also found that the information on the VESC is scattered around the internet. The resources is also sometimes
based on a Debian-based Linux system which adds more work for someone using another distribution of Linux. This could
hinder the implementation facility for new users. We struggled particularly with the Qt packages for positioning and
game pad. We would therefore recommend the use of a Debian-based Linux system for the computer working with the VESC for the MAD associates.
% ************************************ DYNAMIC MODELLING ***************************************************************
\section{Dynamic Modelling and Control of the Bicycle-Cargo System}
\subsection{Dynamic System Modelling}
The studied system consists of a bicycle towing a cargo cart through a rigid mechanical linkage. This link is only used
for steering guidance and does not contribute to the traction force. The main objective is to ensure that the rider
perceives minimal additional effort, such that the overall behaviour remains similar to riding a standard bicycle.
From a control perspective, the rider provides a reference motion in terms of speed and position, while the cargo cart
is expected to follow this reference with minimal delay. The position error between the bicycle and the cargo cart is
computed using a distance sensor, which provides feedback relative to an equilibrium state.
The cargo cart is modelled as the plant of the system. Its rotational dynamics are described using the fundamental
equation of rotational motion:
\begin{equation*}
\sum \tau = J_{\Delta} \times \dot{\omega}
\end{equation*}
where $\tau$ is the total torque applied to the system, $J_{\Delta}$ is the equivalent moment of inertia, and $\omega$
is the angular velocity.
The total torque is composed of the motor torque $\tau_m$ and a friction torque modelled as:
\begin{equation*}
\tau_f = -f \times \omega
\end{equation*}
where $f$ is the viscous friction coefficient.
The resulting dynamic equation becomes:
\begin{equation*}
J_{\Delta} \dot{\omega} = \tau_m - f \omega
\end{equation*}
In the Laplace domain, this leads to:
\begin{equation*}
\omega(s) = \frac{\tau_m(s)}{J_{\Delta} s + f}
\end{equation*}
Since the linear velocity is related to angular velocity by the wheel radius $R$, we obtain:
\begin{equation*}
v(s) = R \times \omega(s)
\end{equation*}
Thus, the transfer function between motor torque and linear velocity is:
\begin{equation*}
\frac{v(s)}{\tau_m(s)} = \frac{R}{J_{\Delta} s + f}
\end{equation*}
\subsection{PI-Based Control Strategy}
\label{subsec:Simulink_model}
Based on this model, a Simulink representation of the system was developed. The controlled system includes a feedback
loop using a PI (Proportional-Integral) controller in order to regulate the position error between the bicycle and the cargo cart.
Since the reference input is a ramp signal (representing the bicycle position over time), an integral action is required
to ensure zero steady-state error and accurate tracking of the reference trajectory.
The closed-loop Simulink model of the system is shown in Fig.~\ref{fig:simulink-closedloop}.
The control error is defined as the difference between a desired relative position and the measured displacement between
the bicycle and the cargo cart:
\begin{equation*}
e(t) = e_{\text{ref}} - (x_{\text{bike}} - x_{\text{cart}})
\end{equation*}
where $e_{\text{ref}} = \SI{-0.5}{\meter}$ represents the desired equilibrium offset between both systems.
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{./Figures/sys_dyn_matlab.png}
\caption{Closed-loop model of the bicycle-cargo system with PI control.}
\label{fig:simulink-closedloop}
\end{figure}
\subsection{Control Architecture Exploration}
Beyond the standard PI control structure simulated in the previous sections, this project explores a more sophisticated
control law: the Cascaded Loop Architecture. This approach is envisioned as a high-level software enhancement to meet
the robustness and safety requirements inherent to electric cargo mobility. This means that our control law includes two
feedback loops that use two different physical parameters.
The selected cascaded structure is a well-established industry standard, particularly in high-performance motion control
and robotics. Similar architectures are widely employed in Automated Guided Vehicles (AGVs) and platooning systems,
where a “follower” unit must synchronize its dynamics with a “leader” unit through precise feedback loops.
The proposed approach decomposes the complex task of “cart following” into manageable sub-tasks by nesting control
loops:
\begin{itemize}
\item Outer loop: Position control layer.
Using a linear encoder or distance sensor mounted on the trailer’s hitch, the system measures the relative
displacement (error) between the bicycle and the cargo cart. This error is processed by a Proportional (P)
Controller. The primary goal of this stage is to translate physical distance into a target velocity setpoint. By
saturating the output of this loop, we can prevent the cargo cart from ever exceeding the bicycle’s speed, thereby
ensuring it never “pushes” the cyclist.
\item Inner loop: Velocity control layer.
The velocity setpoint generated by the outer loop is fed into this internal layer, the inner loop is responsible for
commanding the motor torque directly to compensate for immediate mechanical disturbances. Because this loop operates
at a higher frequency, it can reject disturbances such as sudden changes in rolling resistance or friction, before
they significantly impact the overall position error.
\end{itemize}
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{./Figures/Schema_Autom_PIR.pdf}
\caption{Cascaded control architecture for the bicycle-cargo system.}
\label{fig:cascaded-loop}
\end{figure}
Fig.~\ref{fig:cascaded-loop} illustrates the cascade architecture for the dynamics of the cargo cart.
In this scheme, $x_{\text{ref}}$ denotes the desired position of the bicycle, whereas $x$ represents the measured
position of the cargo cart.
An outer Proportional Controller $K_{p,x}$ converts the position error $e_x = x_{\text{ref}} - x$ into a velocity
reference $v_{\text{ref}}$, which serves as the set-point for the inner loop.
The inner Proportional Controller $K_{p,v}$ then transforms the velocity error $e_v = v_{\text{ref}} - v$ into a
torque reference $\tau_{\text{ref}}$.
This command is processed by the motor/actuator block, which delivers the actual torque $\tau$.
The model of the plant maps this torque to velocity $v$, and the integrator $1/s$ reconstructs the position $x$.
The adoption of a cascaded loop architecture offers decisive advantages but comes with disadvantages. This precision
introduces increased complexity: the multiplication of tuning parameters and the requirement for high-resolution
feedback sensors, such as encoders, raise hardware costs and must come with high-performance software. These technical
constraints represent a significant challenge and may raise other issues.
% ******************************** RESULTS **************************************************************************
\section{Results}
\subsection{FOC Controller Validation}
\subsubsection{Current Status Summary}
Table~\ref{tab:foc_status} summarizes the current status of the FOC
controller development.
\begin{table}[htbp]
\caption{FOC controller development status}
\label{tab:foc_status}
\centering
\begin{tabular}{l c}
\toprule
\textbf{Task} & \textbf{Status} \\
\midrule
VESC firmware compilation & Completed \\
Pin compatibility (F405 / L476) & Completed \\
Schematic design (KiCad) & Completed \\
ERC validation & Completed \\
PCB routing & In progress \\
Tile footprint correction & In progress \\
Board manufacturing & Planned \\
Hardware testing & Planned \\
\bottomrule
\end{tabular}
\end{table}
\subsection{Bicycle-Cargo System Control Results}
\subsubsection{Simulation Results}
The closed-loop Simulink model presented in the subsection~\ref{subsec:Simulink_model} was used to evaluate the
performance of the proposed PI-based (Proportional-Integral) control strategy.
Figure~\ref{fig:tracking-error} shows the evolution of the tracking error between the bicycle and the cargo cart during
simulation. The response exhibits an initial transient phase followed by a progressive convergence toward the desired
equilibrium position, demonstrating stable closed-loop behaviour and satisfactory tracking performance.
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{./Figures/error_fig.png}
\caption{Position tracking error between bicycle and cargo cart.}
\label{fig:tracking-error}
\end{figure}
\subsubsection{Experimental Load Characterization}
Experimental tests were conducted on flat terrain in order to evaluate the influence of mechanical load on the motor
current consumption of the cargo cart system. The system was powered using a \SI{48}{\volt} battery pack.
Current measurements were acquired using an Analog Discovery 2 connected to a computer running the WaveForms software
environment. A current clamp probe was used to measure the motor current, and the signals were sampled at
\SI{1}{\kilo\hertz}.
During each test, the throttle command was set to its maximum value in order to produce the highest possible
acceleration. Once the maximum speed was reached, the motor current naturally decreased and stabilised as the motor
only compensated for rolling resistance and friction effects.
Three loading conditions were investigated corresponding approximately to one, two, and three passengers inside the
cargo cart. The motor current measured during these experiments is shown in Fig.~\ref{fig:motor-currents}.
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{./Figures/Motor_currents.pdf}
\caption{Measured motor current under three loading conditions.}
\label{fig:motor-currents}
\end{figure}
The results show a significant current peak during the acceleration phase, reaching the controller limit of
approximately \SI{25}{\ampere}. After this transient phase, the current decreases and converges toward a lower
steady-state value corresponding mainly to friction and resistive force compensation.
As expected, higher loading conditions resulted in higher steady-state current consumption, indicating an increase in
the required motor torque. In addition, the duration during which the current remained close to the maximum controller
limit also increased with heavier loads, reflecting the longer acceleration time required to reach steady-state
operation.
These variations are mainly attributed to terrain irregularities, throttle response fluctuations, and limitations
associated with the measurement setup and current probe acquisition chain.
However, due to the absence of direct velocity measurements during the experiments, only qualitative observations could
be extracted from these tests. Consequently, a precise estimation of dynamic friction parameters and energy efficiency
could not be achieved.
\subsection{FOC Controller Validation}
\subsubsection{Current Status Summary}
Table~\ref{tab:foc_status} summarizes the current status of the FOC
controller development.
\begin{table}[htbp]
\caption{FOC controller development status}
\label{tab:foc_status}
\centering
\begin{tabular}{l c}
\toprule
\textbf{Task} & \textbf{Status} \\
\midrule
VESC firmware compilation & Completed \\
Pin compatibility (F405 / L476) & Completed \\
Schematic design (KiCad) & Completed \\
ERC validation & Completed \\
PCB routing & In progress \\
Tile footprint correction & In progress \\
Board manufacturing & Planned \\
Hardware testing & Planned \\
\bottomrule
\end{tabular}
\end{table}
% ******************************** DISCUSSION **************************************************************************
\section{Discussion}
This project could be seen as an introduction to the VESC project for someone who don't know about it from beforehand,
the challenges the new users face during setup, as well as a demand for clear expectations concerning documentation
on the subject. The project the MAD is leading should probably not be a fork of the project, as the project is still
in development.
As a final note, this proved to be a project which could easily be developed into several different projects in
different fields. Some projects could be continued later on as a different PIR subject, other could be proposed to later years in different specialisations like TLS-SEC, ESPE. Our thoughts on the following projects that could be
The fabrication line for electronics is globalised. This is okay in a stable world, but it could be a problem in a world
full of instability, be it war, blockages, or tariffs. The idea of opening a specialisation in cooperation with AIME came
up as an idea.
For TLS-SEC the subject could be the design for a fitting mechanism to restrict certain privileges to certified
personnel that could be used in the C programming language. Later down the line we could also see the possibility to
analyse the Bluetooth frames in order to manipulate them in order to change important parameters.
The continuation on the PCB could be a subject fitting an ESPE specialisation.
The proposition of and supply of a VESC system to play with and troubleshoot could be a good rule of thumb, which
allows for a quicker start and gives among other things an idea of the budget and the supply line used by a entity
in the sector. Proposing a visit could also be one way to familiarise students with the association.
What should be a clear conclusion from our test with the jammer is that a controller based on Bluetooth alone should be
avoided when possible and practical. Examples where this could be relevant include electric skateboards, as cables could
impose a tripping hazard. There, an encapsulation of an encrypted control frame could be an thought.
% ******************************** Perspectives and Future Work ************************************************************
\section{Perspectives and Future Work}
Based on the results obtained and the limitations identified during this
project, several directions for future work are proposed.
\subsection{Hardware Completion and Testing}
The VESC-based FOC PCB requires routing completion and prototype
manufacturing. Once fabricated, the board must be tested under real
operating conditions (varying loads, road profiles, and battery voltage).
\subsection{Control Strategy Enhancement}
Future work includes the implementation and comparison of the PI controller and the cascaded control structure on the
cargo cart, in order to evaluate their performance under different conditions (load variations, road profiles, etc.).
In addition, the lack of direct velocity measurements limited the ability to fully characterise the system dynamics.
Adding a speed measurement or improving state estimation would allow a more complete analysis of the model.
% ******************************** CONCLUSION **************************************************************************
\section{Conclusion/Summary}
%\begin{figure}[htbp]
%%\centerline{\includegraphics{fig1.png}}
%caption{Example of a figure caption.}
%\label{fig}
%\end{figure}
%Figure Labels: Use 8 point Times New Roman for Figure labels. Use words
%rather than symbols or abbreviations when writing Figure axis labels to
%avoid confusing the reader. As an example, write the quantity
%``Magnetization'', or ``Magnetization, M'', not just ``M''. If including
%units in the label, present them within parentheses. Do not label axes only
%with units. In the example, write ``Magnetization (A/m)'' or ``Magnetization
%\{A[m(1)]\}'', not just ``A/m''. Do not label axes with a ratio of
%quantities and units. For example, write ``Temperature (K)'', not
%``Temperature/K''.
% ******************************** REMERCIEMENTS **************************************************************************
\section*{Acknowledgment}
The authors would like to thank Pascal Acco and Thierry Rocacher for their
continuous technical guidance and support throughout this project. Their
expertise in power electronics, embedded systems, and PCB design was
invaluable.
We also thank La Manufacture Autonome Décentralisée (LaMAD) for providing
the use case, the technical requirements, and the cargo bike platform used
for validation.
Finally, we acknowledge the INSA Toulouse GEI department for providing
access to laboratory facilities, measurement equipment, and the necessary
components for prototyping.
This work was carried out as part of the 4th-year research project (PIR)
at INSA Toulouse.
% ******************************** IA **************************************************************************
\section*{Statement on AI Usage}
The authors acknowledge the use of generative AI tools during this project, both for the development work and for
writing this paper.
AI was used as a helper in several parts of the project. This includes support for understanding and structuring
technical ideas, and giving suggestions during the development of different subsystems. It was also used to help with
writing, rephrasing, and improving clarity in the report.
However, all final decisions, implementations, and validations were done by the authors. The AI outputs were always
checked, corrected when needed, and adapted based on reliable technical sources and our own experiments and
understanding of the system.
We consider AI as a useful tool to speed up thinking and writing, but not as a source of final technical truth.
Everything related to design choices, analysis, and results was verified and fully controlled by the authors.
The use of AI tools in this work follows the IEEE guidelines for generative AI usage in publications.
%% On découpe le projet pour faciliter le travail en commun
% Début
\input{intro.tex}
\input{related.tex}
\input{gap.tex}
\input{aim.tex}
% Nos parties
\input{foc.tex}
\input{lowtech.tex}
\input{logiciel.tex}
\input{modelisation.tex}
% Fin
\input{resultats.tex}
\input{discussion.tex}
\input{futur.tex}
\input{conclusion.tex}
\input{remerciements.tex}
\input{IA.tex}
%Please number citations consecutively within brackets \cite{b1}. The
%sentence punctuation follows the bracket \cite{b2}. Refer simply to the reference

104
Article_Scientifique/modelisation.tex Normal file
View file

@ -0,0 +1,104 @@
\section{Dynamic Modelling and Control of the Bicycle-Cargo System}
\subsection{Dynamic System Modelling}
The studied system consists of a bicycle towing a cargo cart through a rigid mechanical linkage. This link is only used for steering guidance and does not contribute to the traction force. The main objective is to ensure that the rider perceives minimal additional effort, such that the overall behaviour remains similar to riding a standard bicycle.
From a control perspective, the rider provides a reference motion in terms of speed and position, while the cargo cart is expected to follow this reference with minimal delay. The position error between the bicycle and the cargo cart is computed using a distance sensor, which provides feedback relative to an equilibrium state.
The cargo cart is modelled as the plant of the system. Its rotational dynamics are described using the fundamental equation of rotational motion:
\begin{equation*}
\sum \tau = J_{\Delta} \times \dot{\omega}
\end{equation*}
where $\tau$ is the total torque applied to the system, $J_{\Delta}$ is the equivalent moment of inertia, and $\omega$ is the angular velocity.
The total torque is composed of the motor torque $\tau_m$ and a friction torque modelled as:
\begin{equation*}
\tau_f = -f \times \omega
\end{equation*}
where $f$ is the viscous friction coefficient.
The resulting dynamic equation becomes:
\begin{equation*}
J_{\Delta} \dot{\omega} = \tau_m - f \omega
\end{equation*}
In the Laplace domain, this leads to:
\begin{equation*}
\omega(s) = \frac{\tau_m(s)}{J_{\Delta} s + f}
\end{equation*}
Since the linear velocity is related to angular velocity by the wheel radius $R$, we obtain:
\begin{equation*}
v(s) = R \times \omega(s)
\end{equation*}
Thus, the transfer function between motor torque and linear velocity is:
\begin{equation*}
\frac{v(s)}{\tau_m(s)} = \frac{R}{J_{\Delta} s + f}
\end{equation*}
\subsection{PI-Based Control Strategy}
\label{subsec:Simulink_model}
Based on this model, a Simulink representation of the system was developed. The controlled system includes a feedback loop using a PI (Proportional-Integral) controller in order to regulate the position error between the bicycle and the cargo cart.
Since the reference input is a ramp signal (representing the bicycle position over time), an integral action is required to ensure zero steady-state error and accurate tracking of the reference trajectory.
The closed-loop Simulink model of the system is shown in Fig.~\ref{fig:simulink-closedloop}.
The control error is defined as the difference between a desired relative position and the measured displacement between the bicycle and the cargo cart:
\begin{equation*}
e(t) = e_{\text{ref}} - (x_{\text{bike}} - x_{\text{cart}})
\end{equation*}
where $e_{\text{ref}} = \SI{-0.5}{\meter}$ represents the desired equilibrium offset between both systems.
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{./Figures/sys_dyn_matlab.png}
\caption{Closed-loop model of the bicycle-cargo system with PI control.}
\label{fig:simulink-closedloop}
\end{figure}
\subsection{Control Architecture Exploration}
Beyond the standard PI control structure simulated in the previous sections, this project explores a more sophisticated control law: the Cascaded Loop Architecture. This approach is envisioned as a high-level software enhancement to meet the robustness and safety requirements inherent to electric cargo mobility. This means that our control law includes two feedback loops that use two different physical parameters.
The selected cascaded structure is a well-established industry standard, particularly in high-performance motion control and robotics. Similar architectures are widely employed in Automated Guided Vehicles (AGVs) and platooning systems, where a “follower” unit must synchronize its dynamics with a “leader” unit through precise feedback loops.
The proposed approach decomposes the complex task of “cart following” into manageable sub-tasks by nesting control loops:
\begin{itemize}
\item Outer loop: Position control layer.
Using a linear encoder or distance sensor mounted on the trailer’s hitch, the system measures the relative displacement (error) between the bicycle and the cargo cart. This error is processed by a Proportional (P) Controller. The primary goal of this stage is to translate physical distance into a target velocity setpoint. By saturating the output of this loop, we can prevent the cargo cart from ever exceeding the bicycle’s speed, thereby ensuring it never “pushes” the cyclist.
\item Inner loop: Velocity control layer.
The velocity setpoint generated by the outer loop is fed into this internal layer, the inner loop is responsible for commanding the motor torque directly to compensate for immediate mechanical disturbances. Because this loop operates at a higher frequency, it can reject disturbances such as sudden changes in rolling resistance or friction, before they significantly impact the overall position error.
\end{itemize}
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{./Figures/Schema_Autom_PIR.pdf}
\caption{Cascaded control architecture for the bicycle-cargo system.}
\label{fig:cascaded-loop}
\end{figure}
Fig.~\ref{fig:cascaded-loop} illustrates the cascade architecture for the dynamics of the cargo cart. In this scheme, $x_{\text{ref}}$ denotes the desired position of the bicycle, whereas $x$ represents the measured position of the cargo cart.
An outer Proportional Controller $K_{p,x}$ converts the position error $e_x = x_{\text{ref}} - x$ into a velocity reference $v_{\text{ref}}$, which serves as the set-point for the inner loop.
The inner Proportional Controller $K_{p,v}$ then transforms the velocity error $e_v = v_{\text{ref}} - v$ into a torque reference $\tau_{\text{ref}}$.
This command is processed by the motor/actuator block, which delivers the actual torque $\tau$.
The model of the plant maps this torque to velocity $v$, and the integrator $1/s$ reconstructs the position $x$.
The adoption of a cascaded loop architecture offers decisive advantages but comes with disadvantages. This precision introduces increased complexity: the multiplication of tuning parameters and the requirement for high-resolution feedback sensors, such as encoders, raise hardware costs and must come with high-performance software. These technical constraints represent a significant challenge and may raise other issues.

33
Article_Scientifique/related.tex Normal file
View file

@ -0,0 +1,33 @@
\section{Related Work}
\label{sec:relatedwork}
\subsection{Modeling of BLDC Motor}
The electromechanical model of a BLDC (Brushless DC) motor is foundational for understanding its behaviour under different control schemes. BLDC motors are categorized by their back-electromotive force (back-EMF) waveform: trapezoidal or sinusoidal. This distinction is crucial, as the trapezoidal shape inherently leads to torque ripple when the supplied phase currents are not perfectly aligned, directly influencing the choice and effectiveness of the control strategy \cite{patil_analysis_2025}. For a BLDC motor with trapezoidal back-EMF, the electromagnetic torque is given by:
\[
T_e = \frac{e_a i_a + e_b i_b + e_c i_c}{\omega_m}
\]
where \( e_x \) is the back-EMF and \( i_x \) is the phase current \cite{li_quantitative_2019}. The classical d-q reference frame model, ideal for sinusoidal machines, is less suitable for trapezoidal BLDC motors because it assumes sinusoidal flux distribution. Phase-variable modelling in the natural (abc) frame is therefore more appropriate, as it directly accounts for the non-sinusoidal, trapezoidal nature of the back-EMF and the associated harmonics \cite{mohammd_taher_new_2021}.
\subsection{Trapezoidal Commutation (Six-Step Control) for BLDC Motors}
% Trapezoidal commutation, or six-step control, uses Hall-effect sensors to synchronize phase current switching every
% 120 electrical degrees.
Trapezoidal commutation, or Six-Step control, uses bipolar conduction, with two motor phases conducting at any time and current commutation occurring every 120 electrical degrees \cite{gieras_modern_2023}. As commutation depends on rotor position, Six-Step control requires either position sensors (e.g. Hall sensors, encoders, or resolvers) or sensor-less estimation based on back-EMF detection or observers \cite{gieras_modern_2023, gasc_conception_2004}. This method is renowned for its simplicity of implementation and low hardware cost \cite{bhatiya_bldc_2024}. It enables effective torque control but introduces significant torque ripple during commutation events, especially under high load \cite{jomsa-nga_torque_2024}. This ripple generates noise, increases mechanical stress, and reduces overall efficiency \cite{mohammd_taher_new_2021}. Although PWM techniques can mitigate this ripple, they do not completely eliminate it \cite{li_quantitative_2019}.
\subsection{Field-Oriented Control (FOC) for BLDC Motors}
FOC is a vector control strategy that decouples the stator flux and torque components. It transforms three-phase currents into orthogonal \( I_d \) and \( I_q \) components, enabling precise torque control and significant ripple reduction \cite{jomsa-nga_torque_2024}. FOC is particularly effective for BLDC motors with sinusoidal back-EMF but can also be applied to trapezoidal back-EMF motors, albeit with less impressive ripple suppression results \cite{li_quantitative_2019}. It requires greater computational power and more precise position sensors (e.g. encoders). Comparative analysis shows that FOC yields a more stable stator current profile and significantly reduces torque variations compared to trapezoidal control \cite{patil_analysis_2025}.
\subsection{Comparative Analysis: FOC vs. Trapezoidal for Light Electric Vehicles}
\subsubsection{Torque Ripple and User Comfort}
Firstly, torque ripple can be reduced for both control methods by selecting appropriate motor parameters, such as the number of stator slots and rotor poles \cite{gasc_conception_2004}.
FOC substantially reduces torque ripple compared to Six-Step control, directly enhancing ride comfort and minimizing vibrations. Experimental results show a torque ripple of \SI{18.38}{\percent} for FOC versus \SI{35.67}{\percent} for Six-Step control at 500~rpm \cite{jomsa-nga_torque_2024}. Commutation torque ripple (CTR), prominent in Six-Step control, can be specifically targeted and mitigated using advanced control techniques like Model Predictive Control (MPC) while retaining the fundamental simplicity of trapezoidal commutation \cite{mohammd_taher_new_2021}.
\subsubsection{Energy Efficiency}
FOC optimizes torque per ampere (MTPA), improving efficiency at low loads. Six-Step control exhibits lower switching losses at high speeds \cite{li_quantitative_2019}.
\subsubsection{Complexity, Cost, and Low-Tech Suitability}
Six-Step control is inherently simpler, cheaper, and more robust, making it a prime candidate for low-tech applications. Research focused on reducing propulsion system costs proposes simplified hardware topologies, such as 4-switch inverters (instead of 6) coupled with direct current control strategies, maintaining acceptable performance while significantly lowering hardware costs \cite{lee_advanced_2001}. FOC, while superior in performance, is more complex to implement and carries higher hardware costs (sensors, processing power).
\subsubsection{Dynamic Response}
FOC provides faster response times and better load disturbance rejection \cite{jomsa-nga_torque_2024}.

11
Article_Scientifique/remerciements.tex Normal file
View file

@ -0,0 +1,11 @@
\section*{Acknowledgment}
The authors would like to thank Pascal Acco and Thierry Rocacher for their continuous technical guidance and support throughout this project. Their expertise in power electronics, embedded systems, and PCB design was invaluable.
We also thank La Manufacture Autonome Décentralisée (LaMAD) for providing the use case, the technical requirements, and the cargo bike platform used for validation.
Finally, we acknowledge the INSA Toulouse GEI department for providing access to laboratory facilities, measurement equipment, and the necessary components for prototyping.
This work was carried out as part of the 4th-year research project (PIR) at INSA Toulouse.

92
Article_Scientifique/resultats.tex Normal file
View file

@ -0,0 +1,92 @@
\section{Results}
\subsection{FOC Controller Validation}
\subsubsection{Current Status Summary}
Table~\ref{tab:foc_status} summarizes the current status of the FOC
controller development.
\begin{table}[htbp]
\caption{FOC controller development status}
\label{tab:foc_status}
\centering
\begin{tabular}{l c}
\toprule
\textbf{Task} & \textbf{Status} \\
\midrule
VESC firmware compilation & Completed \\
Pin compatibility (F405 / L476) & Completed \\
Schematic design (KiCad) & Completed \\
ERC validation & Completed \\
PCB routing & In progress \\
Tile footprint correction & In progress \\
Board manufacturing & Planned \\
Hardware testing & Planned \\
\bottomrule
\end{tabular}
\end{table}
\subsection{Bicycle-Cargo System Control Results}
\subsubsection{Simulation Results}
The closed-loop Simulink model presented in the subsection~\ref{subsec:Simulink_model} was used to evaluate the performance of the proposed PI-based (Proportional-Integral) control strategy.
Figure~\ref{fig:tracking-error} shows the evolution of the tracking error between the bicycle and the cargo cart during simulation. The response exhibits an initial transient phase followed by a progressive convergence toward the desired equilibrium position, demonstrating stable closed-loop behaviour and satisfactory tracking performance.
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{./Figures/error_fig.png}
\caption{Position tracking error between bicycle and cargo cart.}
\label{fig:tracking-error}
\end{figure}
\subsubsection{Experimental Load Characterization}
Experimental tests were conducted on flat terrain in order to evaluate the influence of mechanical load on the motor current consumption of the cargo cart system. The system was powered using a \SI{48}{\volt} battery pack.
Current measurements were acquired using an Analog Discovery 2 connected to a computer running the WaveForms software environment. A current clamp probe was used to measure the motor current, and the signals were sampled at \SI{1}{\kilo\hertz}.
During each test, the throttle command was set to its maximum value in order to produce the highest possible acceleration. Once the maximum speed was reached, the motor current naturally decreased and stabilised as the motor only compensated for rolling resistance and friction effects.
Three loading conditions were investigated corresponding approximately to one, two, and three passengers inside the cargo cart. The motor current measured during these experiments is shown in Fig.~\ref{fig:motor-currents}.
\begin{figure}[!h]
\centering
\includegraphics[width=\linewidth]{./Figures/Motor_currents.pdf}
\caption{Measured motor current under three loading conditions.}
\label{fig:motor-currents}
\end{figure}
The results show a significant current peak during the acceleration phase, reaching the controller limit of approximately \SI{25}{\ampere}. After this transient phase, the current decreases and converges toward a lower steady-state value corresponding mainly to friction and resistive force compensation.
As expected, higher loading conditions resulted in higher steady-state current consumption, indicating an increase in the required motor torque. In addition, the duration during which the current remained close to the maximum controller limit also increased with heavier loads, reflecting the longer acceleration time required to reach steady-state operation.
These variations are mainly attributed to terrain irregularities, throttle response fluctuations, and limitations associated with the measurement setup and current probe acquisition chain.
However, due to the absence of direct velocity measurements during the experiments, only qualitative observations could be extracted from these tests. Consequently, a precise estimation of dynamic friction parameters and energy efficiency could not be achieved.
\subsection{FOC Controller Validation}
\subsubsection{Current Status Summary}
Table~\ref{tab:foc_status} summarizes the current status of the FOC controller development.
\begin{table}[htbp]
\caption{FOC controller development status}
\label{tab:foc_status}
\centering
\begin{tabular}{l c}
\toprule
\textbf{Task} & \textbf{Status} \\
\midrule
VESC firmware compilation & Completed \\
Pin compatibility (F405 / L476) & Completed \\
Schematic design (KiCad) & Completed \\
ERC validation & Completed \\
PCB routing & In progress \\
Tile footprint correction & In progress \\
Board manufacturing & Planned \\
Hardware testing & Planned \\
\bottomrule
\end{tabular}
\end{table}