Current topics

Medical image segmentation emerging as a pivotal tool for enhanced imaging analysis. By employing sophisticated algorithms and machine learning techniques, medical image segmentation enables the precise extraction of anatomical structures from complex imaging data. This process not only enhances the visualization of critical structures but also facilitates accurate diagnosis of abnormalities.

In recent years, deep learning, particularly Convolutional Neural Networks (CNNs), has emerged as a tool in healthcare, particularly in the realm of medical image classification. By using the power of CNNs, healthcare providers can now achieve remarkable levels of precision in identifying abnormalities, ranging from tumors to fractures, across various modalities such as MRI, CT scans, and X-rays. As deep learning continues to evolve, Its application in medical image classification is paving the way for more personalized and effective diagnostic approaches.

The integration of Convolutional Neural Networks (CNNs) into healthcare diagnostics marks a pivotal advancement in medical image analysis. CNNs, inspired by the human visual system, making them ideal for interpreting medical imaging data. By leveraging CNNs, healthcare professionals can achieve more accurate and efficient diagnoses, leading to improved patient outcomes.

Monitoring the health status of patients in a real-time or near real-time basis became a necessity in hospitals and intensive care units (ICUs). This research aims to define and implement a solution for breathing and pulse monitoring using the Eulerian Video Magnification (EVM) for motion magnification. The solution has to be implemented as an online service for real-time or near real-time monitoring of patients. The full design of the tool together with the workflow starting from the state of the art, requirement specification, design, to the implementation and testing should be well-defined.

Analysis of large medical data sets is a very challenging task due to the amount of data. The used data could be collected using different types of sensors; therefore, it has different data units and different data types. The main aim of this research is to define and implement a workflow for medical data analysis using statistical tools such as R and it’s official libraries. The tool could be used in an offline basis to analyze stored data or in a real-time basis to analyze the medical data in a synchronous way. The full design of the tool together with the workflow starting from the state of the art, requirement specification, design, to the implementation and testing should be well-defined.

Manual medical data collection using data forms is a challenging task especially when dealing with many patients. The main aim of this research is to create and test the next version of the cross-platform application for medical data collection and storage. The application needs to store the data in a specified data format (JSON) in order to enable further data analysis. This research aims to create a cross-platform application which enables users to store their own data and/or their patient’s data. The full design of the software together with the workflow starting from the state of the art, requirement specification, design, to the implementation and testing should be well-defined.

Nowadays, machine learning provides different ways for automated techniques for advanced IT tasks such as real-time and fast face detection in mobile devices. It could be used to help to monitor patient’s health status in real-time for online or offline monitoring basis. However, these techniques often have some technical problems regarding the fast changing in lighting conditions and/or fast movement during the processing. This topic is to build a trained AI model for face detection in mobile devices and provide a better way of auto-detection of perfect lighting condition to ensure smooth and better detection results regarding the surrounding environment of the user. The full design of the software together with the workflow starting from the state of the art, requirement specification, design, to the implementation and testing should be well-defined.

Electrocardiogram (ECG) is an effective way to define the overall human health status using the heart activity of the patients in real-time basis. One of the challenges in this field is to have a real-time monitoring system for ECG signals in order to define normal and abnormal ECG signals. This solution helps to detect deteriorating patients in the intensive care units (ICU) to prevent further serious problems. This topic is to provide a solution for ECG signal processing to detect anomalies in the ECG signals using AI for signal processing. The main idea is to collect different types of ECG leads, teach it for an AI model, and then use it to generate the formal shape of every type and then detect anomalies by comparing the built model with the input data. The full design of the software together with the workflow starting from the state of the art, requirement specification, design, to the implementation and testing should be well-defined.

Electrocardiogram (ECG) is one of the measured human bio-signals which is used widely in hospitals in order to define the heart activity in real-time basis and to have an overall look on the human health status in a non-invasive way. Many wearable devices and/or sensors for daily basis use are developed to provide real-time continuous monitoring for patient’s health status. Digital data collected using these sensors/devices are stored in different data formats or standards. Therefore, it is a real challenge to ensure data interoperability and compatibility. This topic is to build an online service-based solution running on a linux server which helps to interchange ECG data between different data formats or standards and to provide an analysis of the usage of this service. The full design of the software together with the workflow starting from the state of the art, requirement specification, design, to the implementation and testing should be well-defined.

The task:

Mobility patterns play a crucial role in designing efficient public transportation systems in big cities. Taxicab routes serve as the fundamental basis for uncovering these patterns. The student will develop a link prediction to give a traffic prediction in these temporal networks.

The problem is quite similar to this study [1], however, they use the “classic” New York dataset, that works with a zone-based resolution. In this work, the student will go one step forward, and propose, how the original raw GPS data can be processed to provide meaningful networks and how this data can be used for prediction purposes. The graph convolutional network structure is also used in ref. [2]. However, the structure of the applied datasets was significantly different. The previously applied dataset was recorded using sensor stations and the models concentrated on the prediction of future traffic speed. However, in our case, the data is recorded by the GPS of the taxis, hence, the data is not recorded in fixed points, but in a dynamic way, which we need to process to obtain the spatial and temporal characteristics. Moreover, the aim is the prediction of the traffic intensity and not the speed of the vehicles.

This integrated methodology allowed us to gain valuable insights into the travel behaviors and preferences of commuters, providing a foundation for designing efficient and targeted public transportation services. To validate the effectiveness of the approach, the student will apply it to GPS data obtained from a taxicab company in Budapest.

The thesis must contain:

• Review of literature (network science)
• Data processing (temporal network analyses, link prediction)
• Development of a Python-based algorithm
• A detailed description of the algorithms and framework and a demonstration of its application

[1] https://www.mdpi.com/1424-8220/22/16/5982

[2] https://arxiv.org/abs/1709.04875

[3] https://aaai.org/papers/11836-deep-multi-view-spatial-temporal-network-for-taxi-demand-prediction/

[4] https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0274779

The task:

The rapid advancement of technology related to Industry 4.0 has brought about a paradigm shift in the way we interact with assets across various domains. This progress has led to the emergence of the concept of a Human Digital Twin (HDT) – a virtual representation of an individual’s cognitive, physiological, psychological, and behavioural characteristics. The HDT has demonstrated potential as a strategic tool for enhancing productivity, safety, and collaboration within the framework of Industry 5.0.

Meanwhile, the growth in the number of automated processes has correspondingly increased the cognitive load on human operators. In response to this challenge, this paper outlines a process for tracking human cognitive load using the galvanic skin response as a physiological marker and proposes a novel method of building a model based on the Human-Asset Administration Shell (HAAS).

The proposed HAAS framework integrates real-time data streams from wearable sensors, user interactions, contextual information, task specifics and environmental condition to deliver a comprehensive understanding of an individual’s cognitive state, physical wellness, and skills set. To validate the effectiveness of the approach, the student will apply it at the Operator 4.0 laboratory.

The thesis must contain:

• Review of literature (Asset Administration Shell, Human Digital Twin)
• Ontology development
• Development of a Python-based framework
• A detailed description of the algorithms and framework and a demonstration of its application

The electronic control units in vehicles communicate with each other using different protocols, forming a complex automotive electronics network. With the increasing connectivity of vehicles to the internet through mobile networks, the vulnerability of these networks has become a growing concern. However, there is currently no widely accepted methodology for assessing vulnerabilities in vehicle electronic systems.

The aim of the thesis is to develop a methodology for vulnerability analysis of networks in vehicles, with the possibility of using the MulVAL attack graph generator application as one of several implementation options.

To address this issue, the student should review existing literature on vulnerability analysis of traditional IP-based networks, which involves identifying all possible attack paths, and apply this approach to the automotive electronics network. The research will involve identifying data sources and collection methods required for vulnerability analysis, generating an attack graph using MulVAL or other suitable applications, and evaluating the results.

Development of a model predictive regulator in a type 1 diabetes patient simulator.

Student’s task:

• To get acquainted with the patient simulator developed at the university and then the further development of it, implementation and testing of optimization procedures.
• Has to carry out the developments in the Python programming language, Advantage: the ability of programming mathematical procedures
• Has to test the controller on a large virtual cohort on an UVA/Padova simulator approved by the FDA for preclinical testing.

The implemented improvements will be used in a mobile phone application.

Basic description of the topic:

Industrial robot arms have extremely good repetition accuracy, but their absolute accuracy, which gives you the accuracy with which a given displacement/rotation can be achieved. This depends on the robot model used for the calculation, which is calibrated by the manufacturers and attempts to track deviations from the nominal model and give a mean value for inaccuracies due to elastic deformations and drive chain uncertainties.

This usually causes 5-10x greater inaccuracy than the robot’s repetition accuracy. This can be improved by improving the robot model. Since the early days of robotics, methods have been developed for this, both for the geometric description of the robot and for the additional uncertainty. Without claiming to be exhaustive, one of them is that the model is not designed for the entire workspace, but only for a narrow zone where precise operations are required. Fuzzy interpolation between these zones is also an exciting issue. Also, adding a soft computing layer to the traditional geometric model to describe additional properties can be a tool for calibration.

The question arose during the handling of biological samples, where the situation is complicated by the fact that, in this case, the glass plates are identified based on the installed camera. The goal is to develop the robot model and the calibration process for more accurate grasping.

Knowledge required for the task:

• C++ and MATLAB (or Python) programming knowledge
• Git version control system
• Basic knowledge of robot modeling (Denavit-Hartenberg parameters)
• Optimization methods

Detailed tasks:

1.  Getting to know the parameters of the Denavit-Hartenberg model and related newer methods (e.g. Hayati), and the calibration methods based on them;
2. Processing of the related literature;
3. Developing a development plan;
4. Data collection with the robotic arm;
5. Implementation, testing and optimization of the calibration procedure using the data;
6. Drawing a conclusion, comparing it with the currently used procedure;
7. Evaluation and publication of results.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

Basic description of the topic:

Complex mechatronic systems increasingly rely on the measurements of sensors and sensor systems. These are usually asynchronous, with different noise, and sometimes with drift, and can measure certain properties of the system. The function of the filter is to provide an optimal estimate of the state of the system based on a model that describes these properties well, for example, the movement state in the case of the localisation task.

In recent decades, the international community has carried out extremely extensive research on the subject. The diploma work identifies, understands and compares the main directions of these methods based on the movement data of the PlatypOUs mobile robot platform developed at the university offline and comparing the efficiency of the procedures with the currently used program packages.

Knowledge required for the task:

• C++ and MATLAB (or Python) programming knowledge
• Git version tracking system
• Control control theory knowledge (state space model, simulation, filtering)
• Knowledge of stochastic calculations (expected value, standard deviation)

Detailed tasks:

1. Learning the basics of linear/extended/unscented Kalman filter methods;
2. Processing the related literature, the selection of the methods to be compared;
3. Development of localisation model(s);
4. Getting to know the platform, recording test data;
5. Implementation, testing and optimisation of screening procedures based on test data;
6. Concluding, comparing it with the currently used method;
7. Best practice implementation as a ROS component;
8. Evaluation and publication of results.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

The use of progressively more complex mathematical models that can account in detail for ventilation mechanics would help the clinician to devise the optimal therapeutic strategy (in terms of Positive End-Expiratory Pressure, Tidal volumes or Peak Inspiratory Pressure) to maximize alveolar recruitment minimizing Ventilation-Induced Lung Injury. This approach is particularly promising when dealing with pulmonary stiffness caused by (relative) cardiac failure.

This work is conducted in collaboration with medical doctors in Budapest and involves both model construction and real-time model identification from on-line patient data.

In the quest for effective pharmacological approaches to tumor control, reduction and elimination, immunological approaches are being evaluated with potentially dramatic success. This research line studies the interaction of immune and tumor cells in-vitro, with slide, Organ-On-Chip or organoid preparations. The work is conducted in collaboration with biologists and aims at building computational models able to represent and explain cell dynamics so as to allow the evaluation and optimization of different therapy schemes in-silico.

Technology now makes available Continuous Glucose Monitors, able to record glycemia for patients at risk in an essentially continuous way over many days or weeks. This single-variable, somewhat erratic signal carries information on the compensation state of the subject wearing the sensor, and might allow the assessment of the efficacy of the lifestyle or therapeutic regimen. However, in order to analyse this signal, advanced mathematical modelling techniques appear necessary, possibly including fractional stochastic differential equations.

This work is exploring largely unknown topics, is leading edge and offers different interesting sides: numerical, analytical, statistical, physiological.

The purpose of this work is to formulate an agent-based model of the evolution of body size and insulin sensitivity inindividuals exposed to mutual cultural influences regarding food intake and physical activity levels. Two basic sub-models will interact, a “longitudinal” physiological and a “transversal” sociological submodel. The model would eventually explain the evolution of diabetes prevalence in populations and allow the assessment of the efficacy of educational campaigns over decades.

This work is developed in collaboration with CNR IRIB Messina Italy.

Smart alarm clock with the application of deep learning algorithms

Neural network based furniture recognizing application

Teaching virtual racers with artificial intelligence

Single machine shop system on local network

Development of a Java-based car rental web application with Angular framework

Web application to support voice-based group work

Developement of a Kotlin-based Android application to control Canon camera wirelessly

Device register and inventory fullstack JS application

Security focused event management web application

Development of sport statistical web application

Comparison of machine learning algorithms via development of a music recommendation system

Automatic recipe creator application

Business Status integration for Specification business object using SAP BTP service integration technologies

Application for Pseudocode Interpretation and Code Analysis

Solving ARC in latent vector spaces

Custom homepage creator

An emotion-based music recommendation application using a neural network

Full-page comic coloring

Home workout application to analyze exercising form

Natural language processing chatbot in hungarian language

Thyroid nodule detection based on ultrasound images with deep convolutional networks

Air traffic management supported by Artifical Intelligence

Face mask detection

VR Driving School

Optimization and representation of a network of food delivery drones in simulation

Real-time event stream processing, monitoring and regulation

Implementation of Scripting Language for 3D Game Engine

Simulation of fluvial erosion

Deep temporal continuous clustering

Sportevent scheduling with the help of combinatorics

Predictive energy management of household-sized small power plants equipped with Li-ion batteries using Artificial Intelligence

Hand gesture recognition for controlling Powerpoint in complex environment

Offside detection system planning and development on lower league football pitches

Classifying abstract paintings with machine learning

Short description

• This research is focused on awareness creation about HPV and cervical cancer starting from smaller to larger communities.
• This research also aims at teaching people safe sex as a lifestyle to avoid being victims of Sexually Transmitted Infections (STIs).
• The researcher would organize small awareness programs about HPV and cervical cancer for interested people to attend.
• The application of statistical methods with be used for data interpretation.

Relevant skills/knowledge (good if the applicant has it, at start):

• SPSS, excel, and R programming are a knowhow.
• Python programming is a plus.
• Research skills.

Expected new skills (what the applicant gains during the task):

• Data analytical skills.
• Paper publication skill.
• Ability to find the level of association between variables.
• Strong analytical skills for data interpretation with the use of SPSS, excel, and R programming.
• Ability to teach about HPV and cervical cancer disease.

Short description

• This research is focused on awareness creation about HPV and cervical cancer starting from smaller to larger communities.
• This research also aims at teaching people safe sex as a lifestyle to avoid being victims of Sexually Transmitted Infections (STIs).
• There are several tasks which are expected by the student working on this research, which includes the development of parametric, nonparametric, and semiparametric statistical models for data interpretation, finding association between risk factors with the use of statistical tests, and providing suggestions for how to prevent the spread of HPV and mitigate high mortality rate of cervical disease.

Relevant skills/knowledge (good if the applicant has it, at start):

• SPSS, excel, and R programming are a knowhow.
• Python programming is a plus.
• Research skills.

Expected new skills (what the applicant gains during the task):

• Data analytical skills.
• Paper publication skill.
• Ability to find the level of association between variables.
• Strong analytical skills for data interpretation with the use of SPSS, excel, and R programming.

Brief description:

Background:

The automation of industrial processes is getting more and more important in the field of meat processing too, however this poses serious technical challenges in several respects as well. Due to food safety regulations and the natural biological diversity of plants and animals, the reliable automation of these processes requires the development of new kinds of “smart” devices (knives, grippers, etc.) that ensure the safe and reliable operation.

Task:

This topic deals with the practical development of a “smart gripper” software designed for meat indsutry automation. The two-finger gripper estimates the closing (gripping) force using two different methods, closed loop force controllers should be built on these sensors. Part of the task is to compare, analyse and test different control methods, taking into account the intended task of the gripper.

Programming languages, environments: Microcontroller programming (ESP8266, C++)

Task details:

• Getting to know the task-specific challenges of meat industry automation
• Mapping of gripping force control methods
• Software development: implementation of different force controllers
• Objective testing and comparison of the implemented controllers, evaluation of results

Brief description:

Background:

The automation of industrial processes is getting more and more important in the field of meat processing too, however this poses serious technical challenges in several respects as well. Due to food safety regulations and the natural biological diversity of plants and animals, the reliable automation of these processes requires the development of new kinds of “smart” devices (knives, grippers, etc.) that ensure the safe and reliable operation.

Task:

This topic deals with the development of the software of a “smart meat gripper”, reliable slip detection has to be achieved based on opticalflow calculation of the video-feed of the endoscope camera placed inside the gripper.  Software development should be optimized for the Raspberry Pi compute module built into the gripper.

Programming languages, environments: ROS, Python, OpenCV

Task details:

• Getting to know the task-specific challenges of meat industry automation
• Mapping opticalflow-based methods
• Planning: selection of appropriate models, methods, libraries, etc.
• Software development
• Development of objective testing methods, testing, evaluation of results

Brief description:

Background:

Robotic Assisted Minimal Invasive Surgery (RAMIS) integrates the accuracy, precision, and high payload of robots with human problem-solving capabilities, opening up new horizons in 21st century surgery. The most common surgical robot, the da Vinci surgical robotic system however poses new challenges for surgeons as well (e.g. lack of force feedback, limited field of vision, etc.) thus in the Antal Bejczy Center of Intelligent Robotics (BARK) we are developing solutions that can potentially help surgeons to overcome these difficulties.

Task:

Tracking the surgeon’s eye movements is useful in several ways (surgical ability assessment, assisting with operating room communication, robot control). The task of this thesiswork is to implement and examine different position control methods of the endoscopic camera of the da Vinci Robot based on gaze tracking.

Software environments:  ROS, Python/C++

Task details:

• Identification of current state-of-the-art solutions
• Familiarize with the eye gaze tracking system and the da Vinci Robot’s control system
• Implementation of different control methods for the endoscope-holding robot arm
• Performance based assessment of the implemented control methods
• User-experience based assessment of the implemented control methods

Manipulating (picking, moving, placing) objects is one of the most dominant functionalities of robots used in industry. In order to manipulate objects, the robot first has to grasp them successfully. In case of bulk production, hard-coded grasping poses can be sufficient, but recently the desire for more-and-more flexible robot systems, capable of rapidly adapting to different unknown objects is increasing. Naturally the hand-crafted grasp poses in this kind of applications are not sufficient, so automatic solutions are sought. The majority of the current grasp detection pipelines consider a predicted probability of the success of the grasp to select the best option. However, the easiest way to grasp an object may not be the best choice for all tasks (e.g.: obstructed best-grasp pose, no collision free place pose for given grasp pose, special required movements such as insertion etc.). A grasp predictor that can consider such task (and robot system) constraints might be better suited for an actual robot application. Moveit Task Constructor provides a flexible framework to implement manipulation tasks for robots, and carry out the motion planning for said tasks. The task constraints, environment and robot setup are also easily accessible using the Moveit functionalities.

The goal of the project is to implement a grasp detector in this framework (likely based on PointNetGPD), that also considers task constraints and robot setup in rating the grasp candidates.

Detailed tasks:

• Implement a grasp detector in the mentioned framework (likely based on PointNetGPD), that also considers task constraints and robot setup in rating the grasp candidates;
• Implement alternative grasp proposal methods for comparison and potentially help with the development of the new grasp pose detection algorithm.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

IN the past decades, teleoperation and telesurgery emerged as an independent research domain, then span out to various application areas. The first remote telesurgery system’s concept was born at NASA around 1973, and after many successful trials, the first ever trans-Atlantic surgical procedure was performed in 2001. With advanced control algorithms, and model predictive control, the stability and transparency of such systems can be assured. The fundamental task of the student is to model and simulate a time-delayed system for telesurgery, implement and test various control algorithms to guarantee the high quality of the teleoperation. Critical parameters of the system should be identified and tuned for optimized performance.

Detailed tasks:

1. Understanding and reviewing the background of large distance teleoperation;
2. Learning current methods for controller design and implementation, identify new possible approaches;
3. Learning into Model Predictive Control
4. Selecting and implementing a specific MPC algorithm
5. Assessment of the simulation, evaluation of the results

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

In the past 30 years, robots first conquered factories, then moved on to service applications. While industrial robots are not allowed to closely work together with robots, human-centered robotics is founded on that very close collaboration. Safety therefore is a critical issue for household, healthcare and other service robotics domains.

During the collaboration and co-working with humans, there are numerous safety aspects to consider. Robot manufacturers developed different protocols and methods to ensure high quality task execution. The global standards and unified approaches are yet missing.

Detailed tasks:

• Analysis of critical human-machine co-working scenarios;
• Objective analysis of existing platforms from the safety point of view focusing on service robots in healthcare;
• Developing an assessment and measurement protocol for safety of human-robot collaboration;
• Evaluating the results in the context of international standards

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

During Minimally Invasive Surgery (MIS), the quality and operations of the endoscopic camera have great importance. Surgeons tend to rely high resolution and stereo endoscopes preferably. Over the past 15 years, the da Vinci Surgical System became the dominant master-slave teleoperational surgical robot globally, and there are over 500 thousand surgeries performed with it annually. The da Vinci Classic at the IROB center has the original, low resolution vision system, thus it requires a major upgrade.

Detailed tasks:

• Learning the camera and vision systems of the da Vinci robot;
• Designing a 3D stereocamera and vision system;
• Implementing and integrating the vision system with the robot and the DVRK research platform, implementation of calibration methods;
• Testing and analyzing the results.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

Basic description of the topic:

Interventions of Robot-Assisted Minimally Invasive Surgery (RAMIS) are performed through small incisions, while the area of operation is viewed on endoscopic camera stream. The appearance of teleoperated robotic systems in the operating room made possible to record the movement of the surgeons’ hands, supporting surgical data science. The gathered data can be utilized in a number of research areas, such as surgical skill assessment or surgical subtask automation.

The task is to plan and execute a trial in phantom environment on the research-enhanced da Vinci Surgical System, da Vinci Research Kit (DVRK) available in the Antal Bejczy Center for Intelligent Robotics, and to acquisitive human motion data.  The low level control of the robot is supported by the modular iRob Surgical Automation Framework (irob-saf).  The task includes the processing of the motion data, extracting features and segmenting elements of the workflow.

Detailed tasks:

• Literature overview in the field of RAMIS and the analysis of surgical motion patterns;
• Get acquainted with ROS, the da Vinci Surgical System, the DVRK platform and irob-saf;
• Design of an experiment and the required phantom environment, preparation of the system to the acquisition of motion data.
• Execution of the experiments, recording of human motion data;
• Processing of the gathered data using methodologies found in the literature, information extraction.
  

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

Basic description of the topic:

During the last three decades, the advancement of surgery was characterized by the spread of Robot-Assisted Minimally Invasive Surgery (RAMIS). In the new technique, interventions can be performed through small incisions, while the area of operation is viewed on endoscopic camera stream. During open surgery and traditional Minimally Invasive Surgery (MIS), it is a common technique to identify lesions or tumors by palpation, since those stiffness usually different than the surrounding tissues. To date, there is no clinical RAMIS system offering haptic feedback to the surgeon. Additionally to helping the surgeon’s work, the gathered stiffness data can benefit different studies of surgical data science.

The task is to develop a palpation probe for the da Vinci Surgical System that enables stiffness analysis of tissues using palpation. The research-enhanced da Vinci Surgical System, da Vinci Research Kit (DVRK) available for the candidate in the Antal Bejczy Center for Intelligent Robotics. The low level control of the robot is supported by the modular iRob Surgical Automation Framework (irob-saf). The task includes the execution of palpation on a silicone phantom containing different stiffness regions, and the validation of the probe.

Detailed tasks:

• Literature overview in the field of RAMIS and RAMIS palpation;
• Get acquainted with ROS, the da Vinci Surgical System, the DVRK platform and irob-saf;
• Design of a palpation probe based on the methods found in the literature;
• Fabrication of the palpation probe, hardware and software integration;
• Integrating the vision system with the robot and the DVRK research platform and irob-saf;
• Execution of palpation in a phantom environment, analysis of the results and validation of the probe.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

Basic description of the topic:

During the last three decades, the rapid spread of Robot-Assisted Minimally Invasive Surgery (RAMIS) induced a revolution to the surgical practice. In the new technique, interventions can be performed through small incisions, while the area of operation is viewed on endoscopic camera stream. Despite its benefits, RAMIS also presented difficulties to the surgeons, some subtasks of the operation became monotonous and time-consuming. A significant portion of current research efforts aims to automate these subtasks to reduce the cognitive load on the surgeon. Although solutions on automating interventions performed on rigid tissues exist, automation regarding soft tissues is still highly challenging. During the last years, the automation of surgical training exercises gained currency as a simplified model of more complex subtasks.

The task is the high level implementation of autonomous surgical training exercises on the research-enhanced da Vinci Surgical System, da Vinci Research Kit (DVRK) available in the Antal Bejczy Center for Intelligent Robotics. The low level control of the robot is supported by the modular iRob Surgical Automation Framework (irob-saf). The task includes object detection and pose estimation on RGB-D camera stream, high level implementation of the workflow of the exercise, and the validation of the solution.

Detailed tasks:

• Literature overview in the field of surgical subtask automation;
• Get acquainted with ROS, the da Vinci Surgical System, the DVRK platform and irob-saf;
• Get acquainted with the used RGB-D camera;
• Object detection and pose estimation;
• Integrating the vision system with the robot and the DVRK research platform and irob-saf;
• High level implementation of the workflow of the chosen surgical training exercise
• Testing and analyzing the results.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

Basic description of the topic:

During the last decades, the rapid spread of Robot-Assisted Minimally Invasive Surgery (RAMIS) induced a revolution to the surgical practice. In the new technique, interventions can be performed through small incisions, while the area of operation is viewed on endoscopic camera stream. Despite its benefits, RAMIS also presented difficulties to the surgeons, some subtasks of the operation became monotonous and time-consuming, A significant portion of current research efforts aims to automate these subtasks to reduce the cognitive load on the surgeon. Although solutions on automating interventions performed on rigid tissues exist, automation regarding soft tissues is still highly challenging. Almost all of the current research projects aim partial automation, thus require constant supervision by the surgeon.

The task is the implementation of an effective Human-Machine Interface (HMI) for partial automation in surgery. The research-enhanced da Vinci Surgical System, da Vinci Research Kit (DVRK) available in the Antal Bejczy Center for Intelligent Robotics. The autonomous execution is supported by Robot Operating Sytem (ROS) based modular iRob Surgical Automation Framework (irob-saf). The task includes the theoretical development of the method of communication between the surgeon and the autonomous system, implementation and integration of the developed solution, and the validation of the system.

Detailed tasks:

• Literature overview in the field of surgical subtask automation and HMIs;
• Get acquainted with ROS, the da Vinci Surgical System, the DVRK platform and irob-saf;
• Development of an effective HMI methodology for partial automation in surgery;
• Integrating the HMI with the robot, the DVRK research platform and irob-saf;
• Execution and evaluation of user studies.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

Basic description of the topic:

The rapid spread of Robot-Assisted Minimally Invasive Surgery (RAMIS) induced a revolution to the surgical practice; a number of interventions can be performed using teleoperated manipulators through small incisions.  The new technique presented new difficulties to the surgeons; some subtasks of the operation became monotonous and time-consuming. Many believe that the next step in the development of surgery will be subtask-level automation. One of the current challenges of surgical subtask automation is to achieve sub-millimeter accuracy using the image guided manipulators.

The task is to overview the literature of hand-eye registration in the field of surgical subtask automation, and to implement a hand-eye registration application based on the found techniques.  The research-enhanced da Vinci Surgical System, da Vinci Research Kit (DVRK) available in the Antal Bejczy Center for Intelligent Robotics. The work is also supported by the Robot Operating Sytem (ROS) based modular iRob Surgical Automation Framework (irob-saf). The task includes systematic measurement and analysis of the accuracy achieved by the developed application.

Detailed tasks:

• Literature overview in the field of surgical subtask automation and hand-eye registration;
• Get acquainted with ROS, the da Vinci Surgical System, the DVRK platform and irob-saf;
• Development of an effective hand-eye registration methodology for partial automation in surgery;
• Integrating the hand-eye registration application with the robot, the DVRK research platform and irob-saf;
• Execution and evaluation of the achieved accuracy.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

Examination and analysis of the correlation between the user and YouTube advertisements

Examination of accuracy of the advertisements on YouTube

Testing and analysis of YouTube advertisements using the blackbox principle

Basic description of the topic:

Mushrooms intended for the fresh market are currently harvested exclusively by hand. As a result, the mushroom industry is highly dependent on human labor. About half of the operating costs of mushroom farms are labor costs, which consist mainly of harvesting costs. Rising wages, stricter consideration of working hours and working conditions also increase the problems of manual labor. For these reasons, there is a great need to automate the harvest. The first step in automation is the detection of individual fungi using machine vision devices, as well as continuous measurement of environmental parameters. Detection refers to the use of a sensing device that can be used to identify mushroom heads that meet the given conditions and to facilitate the harvesting process. The student’s task is to design this system of devices and implement the monitoring system on a small sample model.

Detailed tasks:

• An introduction, a justification for the choice of topic;
• Description of the current technological steps and conditions of mushroom growing;
• A review of the literature on the automation of mushroom cultivation;
• Description of the necessary sensors selected for monitoring the cultivation parameters and the devices for the implementation of machine vision;
• Documentation of hardware and software design of an easy-to-install sensor and camera system taking into account growing conditions;
• Documentation of the design steps and demonstration of the working sample system;
• Summary;
• More development ideas.

Basic description of the topic:

Industrial robot arms have extremely good repetition accuracy, but their absolute accuracy, which gives you the accuracy with which a given displacement/rotation can be achieved. This depends on the robot model used for the calculation, which is calibrated by the manufacturers and attempts to track deviations from the nominal model and give a mean value for inaccuracies due to elastic deformations and drive chain uncertainties.

This usually causes 5-10x greater inaccuracy than the robot’s repetition accuracy. This can be improved by improving the robot model. Since the early days of robotics, methods have been developed for this, both for the geometric description of the robot and for the additional uncertainty. Without claiming to be exhaustive, one of them is that the model is not designed for the entire workspace, but only for a narrow zone where precise operations are required. Fuzzy interpolation between these zones is also an exciting issue. Also, adding a soft computing layer to the traditional geometric model to describe additional properties can be a tool for calibration.

The question arose during the handling of biological samples, where the situation is complicated by the fact that, in this case, the glass plates are identified based on the installed camera. The goal is to develop the robot model and the calibration process for more accurate grasping.

Knowledge required for the task:

• C++ and MATLAB (or Python) programming knowledge
• Git version control system
• Basic knowledge of robot modeling (Denavit-Hartenberg parameters)
• Optimization methods

Detailed tasks:

1.  Getting to know the parameters of the Denavit-Hartenberg model and related newer methods (e.g. Hayati), and the calibration methods based on them;
2. Processing of the related literature;
3. Developing a development plan;
4. Data collection with the robotic arm;
5. Implementation, testing and optimization of the calibration procedure using the data;
6. Drawing a conclusion, comparing it with the currently used procedure;
7. Evaluation and publication of results.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

Basic description of the topic:

Complex mechatronic systems increasingly rely on the measurements of sensors and sensor systems. These are usually asynchronous, with different noise, and sometimes with drift, and can measure certain properties of the system. The function of the filter is to provide an optimal estimate of the state of the system based on a model that describes these properties well, for example, the movement state in the case of the localisation task.

In recent decades, the international community has carried out extremely extensive research on the subject. The diploma work identifies, understands and compares the main directions of these methods based on the movement data of the PlatypOUs mobile robot platform developed at the university offline and comparing the efficiency of the procedures with the currently used program packages.

Knowledge required for the task:

• C++ and MATLAB (or Python) programming knowledge
• Git version tracking system
• Control control theory knowledge (state space model, simulation, filtering)
• Knowledge of stochastic calculations (expected value, standard deviation)

Detailed tasks:

1. Learning the basics of linear/extended/unscented Kalman filter methods;
2. Processing the related literature, the selection of the methods to be compared;
3. Development of localisation model(s);
4. Getting to know the platform, recording test data;
5. Implementation, testing and optimisation of screening procedures based on test data;
6. Concluding, comparing it with the currently used method;
7. Best practice implementation as a ROS component;
8. Evaluation and publication of results.

Over the course of the project, the student will get involved with the various research project of the Antal Bejczy Center for Intelligent Robotics.

While ERP inetgrates all the core processes that are needed to run a company: finance, manufacturing, HR, supply chain, services, procurement, … it still has well defined and (also) separately functioning components. The task is to analyse a company and its functional operations and to identify a certain field that needs IT support. While looking into the operations of the given field students need to understand the underlying tasks and processes and their need for data. Based on this understanding a database plan has to be established and a program has to be created that is able to support the task under scrutiny.

The aim of the development is to create a web-based system that is able to screen preschool children for autism using easily accessible interfaces and tools (e.g. web camera) following a specific assessment mechanism (questionnaire developed by psychologists and interactive game playing).

The students will be responsible for integrating third party components into the web-based autism detection system under development, as well as providing support for the ongoing development of front-end and back-end parts. In addition, they will facilitate the development of the artificial intelligence module to be used for detection.

Front-end: Vue.js

Back-end: JAVA, PhP

AI: under test

The students’ tasks are the development of different functionalities of the system developed by the Centre for Life Cycle Regulation Research at the University of Óbuda
diabetes decision support system.

Main themes:
1) Development of functionalities related to an automated insulin delivery system.
2) Development of functionalities related to blood glucose prediction.
3) Development of automated logging functionalities, back-end integration.
4) Third-party systems, hardware testing, back-end integration.
5) Testing and implementation of mathematical models, back-end integration.
6) Big Data data analysis, processing of different data sources.
7) Development of web/mobile interfaces, components.
8) Development and integration of object-based universal system components.
9) Development of test cases.

Detailing:
1) Development of automatic control functions on model based and model independent ways (modern, robust, adaptive controls, machine intelligence based controls
(reinforcement learning, others)); development of safety functionalities; development of code/system architecture; deploy to cloud computing system.
2) Refinement of blood glucose prediction (model-based) (incorporation of sub-models, testing, consideration of secondary effects (e.g. circadian rhythm)).
3) Development of automations to facilitate diabetic patient diaries (e.g. meal recognition, physical activity recognition).
4) Testing AndroidAPS, OpenAPS systems, isolating and integrating functionalities, integrating activity tracking systems (e.g. GHealth).
5) Investigation, implementation, integration of blood glucose models, HbA1c models, physical activity models, etc., development of automatic identification functionalities.
6) Analysis and processing of clinical and non-clinical measurement data, development and implementation of data collection methods.
7) Development of API access to specific functionalities, display of results in interfaces, diagrams, intervention signals.
8) Development of standard data exchange interfaces, interfaces and universal components (e.g. Runge-Kutta solver, etc.).
9) Development of test cases (e.g. code quality, code refactor, unit tests), analysis of individual components, modules, etc. according to a given set of requirements.

Expected competences:
Python/MATLAB/Julia, basic complex systems knowledge
English language skills
High degree of autonomy, but also ability to work in a team

Topic description:

In the case of web application development, security is the key function to be concerned. Authentication can be defined as the process of verifying someone’s identity by using pre-required details.

The user can be identified by the unique knowledge of some information, e.g. the combination of a username and a password. Another type of method is based on the ownership of some unique item, typically a token or a QR code. A special type of identification methods is based on the unique biometric parameters of the human user.

In the recent years, as one of the trending technologies in security systems, facial recognition got very popular. The idea behind facial recognition is that the human face is considered as a unique identity of an individual, therefore, a camera image could be used for authentication.

In this thesis, the current available solutions for user authentication on the web are collected, categorized and evaluated, with special attention to biometric identification. After reviewing the relevant literature and analyzing similar applications, a novel method of user identification is designed, developed and evaluated.

Topic description:

Deep neural network have shown a great performance in finding correlation between input variables and a target label. Nowadays, models based on deep learning are usually applied for classification problems. In the case of supervised learning, the model learns from labeled training examples: if the number of samples are large, classification accuracy will rise. However, there are cases when there is a large number of categories, with relatively few samples. Humans are able to learn from only a few examples, an interesting field of research is to create machine learning based solutions to create a similar solution.

The key idea behind one-shot learning is to have a system, where prior knowledge is used as a basis of training, and new categories, new classes, new knowledge is added to this base knowledge during training.

The goal of this thesis is to review current state-of-the-art machine learning techniques for one-shot learning and adjacent fields (e. g. zero-shot learning), design and develop a solution for a one-shot learning task.

The task:

Develop a system (consisting of hardware and software components) that is capable of monitoring and configuring a small artificial environment for green household plants. Evaluate the possible communication protocols, devices, and software components, and plan a system where we can monitor the status of the environment (temperature, humidity, soil acidity, etc.) and we can configure how the automatic components (sprinkler, lighting, etc.) work.

After the planning, implement the system in a cloud/IoT environment so that the result of the development can show how these technologies and these hardware components can be used to emphasize sustainability and achieve a reduced carbon footprint for the created artificial environment.

The main drivers of the fast development of image processing and machine vision were the effective learning algorithms in recent years. The task of the students is to investigate the related literature, moreover, to participate in the development of such software solutions which are able to analyze video sources allow the control of different electrical devices (e.g. drones, mobile phones, computers, etc.) by hand and head movements, face gestures or by using eye-tracking. The topic can be introduced in TDK or other scientific conferences.

Tasks:

• Literature review of the given topic;
• Investigation of the related machine learning algorithms;
• Implementing the appropriate algorithms which can be used and/or developing own solutions;
• Develop and realize a system plan;
• Validate and evaluate the results;