Supervisor: Vedant Dave, M.Sc.;
Univ.-Prof. Dr Elmar Rückert
Start date: 15th August 2023

Theoretical difficulty: Mid
Practical difficulty: High

Abstract

The aim of this Thesis is to predict the electricity price for the Hydrogen plants from open-sourced Energy data provided by the European Network of Transmission System Operators (ENTSO-E) [1]. We explore multiple machine learning techniques to achieve this aim. At the end, a standalone GUI is provided, that can be used in the industry with ease. This work was done in collaboration HyCenta Research GmbH.

Further, this thesis seeks to address the following research questions:

• How do different determinants such as the electricity mix (the proportion of energy from various generation sources), in-house generation, and gas prices, influence the cost of electricity?
• Which machine learning approaches/algorithms are most suitable for accurately predicting future electricity price trends, particularly in Austria or other European countries? 
• To what extent does the sensitivity of our model to inputs, like solar and wind energy, affect its overall accuracy and reliability in predicting electricity prices?

Thesis

Deep Neural Energy Price Forecasting for the Hydrogen Industry

Tentative Work Plan

To achieve the objectives, the following concrete tasks will be focused on:

• Literature review
• Evaluation of SOTA methods
• Designing network and hyperparameter tuning
• Evaluation on unseen country’s data
• Development of Standalone GUI

Related Work

[1]  Hirth, Lion & Mühlenpfordt, Jonathan & Bulkeley, Marisa, 2018. “The ENTSO-E Transparency Platform – A review of Europe’s most ambitious electricity data platform,” Applied Energy, Elsevier, vol. 225(C), pages 1054-1067.

Supervisor: Fotios Lygerakis and Prof. Elmar Rueckert

Start Date: 1st March 2023

Theoretical difficulty: low
Practical difficulty: mid

Abstract

As the interaction with robots becomes an integral part of our daily lives, there is an escalating need for more human-like communication methods with these machines. This surge in robotic integration demands innovative approaches to ensure seamless and intuitive communication. Incorporating sign language, a powerful and unique form of communication predominantly used by the deaf and hard-of-hearing community, can be a pivotal step in this direction. 

By doing so, we not only provide an inclusive and accessible mode of interaction but also establish a non-verbal and non-intrusive way for everyone to engage with robots. This evolution in human-robot interaction will undoubtedly pave the way for more holistic and natural engagements in the future.

Thesis

ROS2-based Human-Robot Interaction Framework with Sign Language

Project Description

The implementation of sign language in human-robot interaction will not only improve the user experience but will also advance the field of robotics and artificial intelligence.

This project will encompass 4 crucial elements.

1. Human Gesture Recognition with CNNs and/or Transformers – Recognizing human gestures in sign language through the development of deep learning methods utilizing a camera.
   • Letter-level
   • Word/Gloss-level
2. Chat Agent with Large Language Models (LLMs) – Developing a gloss chat agent.
3. Finger Spelling/Gloss gesture with Robot Hand/Arm-Hand –
   
   • Human Gesture Imitation
   • Behavior Cloning
   • Offline Reinforcement Learning
4. Software Engineering – Create a seamless human-robot interaction framework using sign language.
   • Develop a ROS-2 framework
   • Develop a robot digital twin on simulation
5. Human-Robot Interaction Evaluation – Evaluate and adopt the more human-like methods for more human-like interaction with a robotic signer.

Supervisor: Univ.-Prof. Dr Elmar Rückert, Niko Feith
Start date: 1st March 2023

Theoretical difficulty: low
Practical difficulty: mid

Abstract

The FANUC CRX-10iA robot arm is a compliant system that can be used for collaborative human-robot tasks. 

In this Bachelor thesis, the abilities of the robot arm are evaluated for such co-worker scenarios. In particular, the reachable space, the robustness of the inverse kinematics, the ability to simulate the system in real-time, and the precision and reliability of the system are analyzed.

To embed the system in our CPS Hub, a ROS2 interface will be developed und used for all experiments. The interface can be used to control the system or to send end receive commands from simulation tools like CoppeliaSim or Gazebo. 

Tentative Work Plan

To achieve our objective, the following concrete tasks will be focused on:

• Literature research
• ROS2 interface implementation for the simulation tool
• ROS2 interface implementation for the real system
• Identification of quality measures and definition of the experiments
• Evaluation 
• Thesis writing

Thesis

Development of a ROS2 Interface for the FANUC CRX-10iA robot arm

Supervisor: Linus Nwankwo, M.Sc.;
Univ.-Prof. Dr Elmar Rückert
Start date: 5th September 2022

Theoretical difficulty: mid
Practical difficulty: mid

Abstract

When observing animals in nature, navigation and walking seem like medial-side tasks. However, training robots to effectively achieve the same objective is still a challenging problem for roboticists and researchers. We aim to autonomously perform tasks like navigating traffic, avoiding obstacles, finding optimal routes, surveying human hazardous areas, etc with a quadrupedal robot. These tasks are useful in commercial, industrial, and military settings, including self-driving cars, warehouse stacking robots, container transport vehicles in ports, and load-bearing companions for military operations.

For over 20 years today, the SLAM approach has been widely used to achieve autonomous navigation, obstacle avoidance, and path planning objectives. SLAM is a crucial problem in robotics, where a robot navigates through an unknown environment while simultaneously creating a map of it. The SLAM problem is challenging as it requires the robot to estimate its pose (position and orientation) relative to the environment and simultaneously estimate the location of landmarks in the environment.

Some of the most common challenges with SLAM are the accumulation of localization errors over time, inaccurate pose estimation on a map, loop closure, etc. These problems have been partly overcome by using Pose Graphs for localization errors, Extended Kalman filters and Monte Carlos localization for pose estimation.

Quadrupedal robots are well-suited for challenging environments, where the surface conditions are non-uniform, e.g. in off-road environments or in warehouses where stairs or obstacles have to be overcome but have the difficulty of non-uniform dynamic movement which poses additional difficulty for SLAM. 

In the context of this thesis, we propose to study the concept of SLAM with its associated algorithms and apply it to a quadrupedal robot (Unitree Go1). Our goal is to provide the robot with certain tasks and commands that it will then have to autonomously execute. For example, navigate rooms, avoid slow-moving objects, follow an object (person), etc.

Tentative Work Plan

To achieve our objective, the following concrete tasks will be focused on:

• Study the concept of SLAM as well as its application in quadrupedal robots.

• Setup and familiarize with the simulation environment:
  • Build the robot model (URDF) for the simulation (optional if you wish to use the existing one)
  • Setup the ROS framework for the simulation (Gazebo, Rviz)
  • Recommended programming tools: C++, Python, Matlab
  •  
• Develop a novel SLAM system for a quadrupedal robot to navigate and map challenging real-world environments:

  • 2D/3D mapping in complex indoor/outdoor environments

  • Localization using either Monte Carlo or extended Kalman filter

  • Establish a path-planning algorithm

• Intermediate presentation:
  • Presenting the results of the literature study
  • Possibility to ask questions about the theoretical background
  • Detailed planning of the next steps
  •  

• Implementation:

  • Simulate the achieved results in a virtual environment (Gazebo, Rviz, etc.)

  • Real-time testing on Unitree Go1 quadrupedal robot.

• Evaluate the performance in various challenging real-world environments, including outdoor terrains, urban environments, and indoor environments with complex structures.
• B.Sc. thesis writing.
• Research paper writing (optional)

Related Work

[1]  Wolfram Burgard, Cyrill Stachniss, Kai Arras, and Maren Bennewitz , ‘SLAM: Simultaneous
Localization and Mapping’,  http://ais.informatik.uni-freiburg.de/teaching/ss12/robotics/slides/12-slam.pdf

[2]  V.Barrile, G. Candela, A. Fotia, ‘Point cloud segmentation using image processing techniques for structural analysis’, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019 

[3]  Łukasz Sobczak , Katarzyna Filus , Adam Domanski and Joanna Domanska, ‘LiDAR Point Cloud Generation for SLAM Algorithm Evaluation’, Sensors 2021, 21, 3313. https://doi.org/10.3390/ s21103313.

Find more about the thesis at this address:

https://cloud.cps.unileoben.ac.at/index.php/s/RybJssqpq68KDYF

Supervisor: DI Nikolaus Feith;
Univ.-Prof. Dr Elmar Rückert
Start date: 15th Februar 2023

Theoretical difficulty: mid
Practical difficulty: mid

Abstract

In the last decade, MotoGP has taken data analytics and telemetrics to a whole new level which has aided in the development and manufacturing of the motorcycles prototypes and their performance on the track.

However, as in other high-end motorsports competitions such as F1, the technological gap keeps getting smaller and, with so, the real potential advantage is gained during testing by trying to find the optimal setup for that specific track and weather conditions. Due to regulations, time dedicated to testing on track is quite scarce, therefore teams and pilots have to find their way around new prototype setups every week to optimize and extract the best performance of the bike.

The main goal of this project is to develop a Bayesian Optimization algorithm that can aid in the fine tuning of ECU parameters of the motorcycle (fuel injection timing and spark ignition timing) while providing a framework and workflow for this methodology.

The work is done in collaboration with the Chair of Cyber-Physical-Systems at Montanuniversitaet Leoben, and Montan Factory Racing, participant of the VII Edition Motostudent Petrol.

Tentative Work Plan

To achieve our objective, the following concrete tasks will be focused on:

• Literature research
• Engine and controller model development in Simulink
• Bayesian optimization algorithm development
• Hardware integration
• Testing in practical applications
• M.Sc. thesis writing and documentation

Related Work

[R1] Isermann, R., Hafner, M. (2001). Mechatronic Combustion Engines – from Modeling to Optimal Control.

[R2] Schillinger, M., Hartmann, B., Skalecki, P., Meister, M., Nguyen-Tuong, D., & Nelles, O. (2017). Safe active learning and safe Bayesian optimization for tuning a PI-controller. IFAC-PapersOnLine, 50(1), 5967-5972.

[R3] Isermann, R. (2014). Engine modeling and control. Berlin: Springers Berlin Heidelberg, 1017.

[R4] Gerhardt, J., Hönninger, H., & Bischof, H. (1998). A new approach to functional and software structure for engine management systems-BOSCH ME7. SAE transactions, 1173-1184.

Thesis

Bayesian Optimization for learning optimal parameters of Electronic Control Units (ECU’s) for Motorcycles

Supervisor: DI Nikolaus Feith;
Konrad Bartsch;
Univ.-Prof. Dr Elmar Rückert
Start date: 8th Februar 2023

Theoretical difficulty: low
Practical difficulty: high

Abstract

Modern IoT devices are powerful elements in complex Cyber-Physical-Systems (CPS). 

However, communicating with such microcontrollers can be challenging and often requires custom software and hardware interfaces. When working with many different devices, this can quickly become overwhelming. 

The goal of this thesis is to develop a generic hardware interface for the ESP32 microcontroller.

Individual hardware devices, sensors, and actuators can be integrated into a CPS through configuration files. Adjusting these files does not require in-depth hardware or software knowledge and allows rapid IoT development and integration via ROS 2.

The power of the generic ROS2 device interface is demonstrated in multiple use cases, e.g., the sensor glove with flex sensors, vibration motors and an IMU, or an ODrive motor controller board for mobile robots. 

Tentative Work Plan

To achieve our objective, the following concrete tasks will be focused on:

• Assess the hardware and software requirements for the interfaces.
• Literature research on related or existing generic ROS2 solutions.
• Development of the generic software program. 
• Use case evaluation of the interface for various devices. Assessment of the performance and limitations. 
• Software documentation in the wiki of the git repository.

• B.Sc. thesis writing
• Research paper contribution with figures, results (optional).

Related Work

[R1] Dauphin, L., Baccelli, E., & Adjih, C. (2018, September). RIOT-ROS2: low-cost robots in IoT controlled via information-centric networking. In 2018 IFIP/IEEE International Conference on Performance Evaluation and Modeling in Wired and Wireless Networks (PEMWN) (pp. 1-6). IEEE.

[R2] Barciś, M., Barciś, A., Tsiogkas, N., & Hellwagner, H. (2021). Information Distribution in Multi-Robot Systems: Generic, Utility-Aware Optimization Middleware. Frontiers in Robotics and AI, 8, 685105.

[R3] Jo, W., Kim, J., Wang, R., Pan, J., Senthilkumaran, R. K., & Min, B. C. (2022). Smartmbot: A ros2-based low-cost and open-source mobile robot platform. arXiv preprint arXiv:2203.08903.

Tutorials and Documentations

[1] ESP32 Tutorials, last visited 09.02.2023, https://randomnerdtutorials.com/getting-started-with-esp32/

[2] ESP32 Tutorials, last visited 09.02.2023, https://www.az-delivery.de/en/blogs/azdelivery-blog-fur-arduino-und-raspberry-pi/esp32-das-multitalent

[3] MAC OS Serial Driver, last visited 09.02.2023, https://github.com/adrianmihalko/ch340g-ch34g-ch34x-mac-os-x-driver

[4] ESP32 Datasheet, last visited 09.02.2023, https://www.espressif.com/sites/default/files/documentation/esp32_datasheet_en.pdf

[5] ROS2 Documentation, last visited 09.02.2023, https://docs.ros.org/en/humble

Thesis

Development of a generic ROS2 Device Interface based on Micro-ROS on a ESP32

Supervisor: Vedant Dave, M.Sc; Univ.-Prof. Dr Elmar Rückert
Start date: 1st May 2022                          Finsihed: 18th October 2022

Theoretical difficulty: Mid
Practical difficulty: Mid

Abstract

Tactile perception is one of the basic senses in humans that utilize almost at every instance. We predict the touch of the object even before touching it, only through vision. If a novel object is encountered, we predict the tactile sensation even before touching. The goal of this project is to predict tactile response that would be experienced if this grasp were performed on the object. This is achieved by extracting the features of the visual data and the tactile information and then learning the mapping between those features. 

We use Intel RealSense depth camera D435i for capturing images of the objects and Seed RH8D Hand with tactile sensors to capture the tactile data in real time(15 dimensional data). The main objective is to perform well on the novel object which have some shared feature representation of the previously seen objects.

Plan

• Literature Research
• Architecture Development
• Dataset Collection from Real Robot.
• Application in Real Robot.
• Master Thesis Writing
• Research Paper Writing (Optional)

Related Work

[1] B. S. Zapata-Impata, P. Gil, Y. Mezouar and F. Torres, “Generation of Tactile Data From 3D Vision and Target Robotic Grasps,” in IEEE Transactions on Haptics, vol. 14, no. 1, pp. 57-67, 1 Jan.-March 2021, doi: 10.1109/TOH.2020.3011899.

[2] Z. Abderrahmane, G. Ganesh, A. Crosnier and A. Cherubini, “A Deep Learning Framework for Tactile Recognition of Known as Well as Novel Objects,” in IEEE Transactions on Industrial Informatics, vol. 16, no. 1, pp. 423-432, Jan. 2020, doi: 10.1109/TII.2019.2898264.

Thesis Document

A Framework for Learning Visual and Tactile Correlation

Supervisor: Univ.-Prof. Dr Elmar Rückert
Start date: 1st of July 2021

Theoretical difficulty: mid
Practical difficulty: mid

Abstract

 A central problem in robotics is the description of the movement of a robot. This task is complex, especially for robots with high degrees of freedom. In the case of complex movements, they can no longer be programmed manually. Instead, they are taught to the robot utilizing machine learning. The Motor Control Learning framework presents an easy-to-use method for generating complex trajectories. Dynamic Movement Primitives is a method for describing movements as a non-linear dynamic system. Here, the trajectories are modelled by weighted basis functions, whereby the machine learning algorithms must determine only the respective weights. Thus, it is possible for complex movements to be defined by a few parameters. As a result, two motion learning methods were implemented. When imitating motion demonstrations, the weights are determined using regression methods. A reinforcement learning algorithm is used for policy optimization to generate waypoint trajectories. For this purpose, the weights are improved iteratively through a cost function using the covariance matrix adaptation evolution strategy. The generated trajectories were evaluated in experiments. 

Thesis Document

A Motor Control Learning Framework for Cyber-Physical-Systems

Supervisors: Univ.-Prof. Dr Elmar Rückert,
Vedant Dave, M.Sc. 
Dr. Christoph Sorger und Dr. Luca Moderer
Univ.-Prof. Martin Stockinger (Chair of Metal Forming)
Start date: ASAP from June 2022

Theoretical difficulty: low
Practical difficulty: low

Abstract

In this thesis, the student has the unique opportunity to investigate supervised machine learning methods for predicting yield strengths using probabilistic regression models and deep learning approaches. The thesis is implemented with support of the MSC Software GmbH and the Stahl- und Walzwerk Marienhütte GmbH in Graz.

In the image above and below you see the production line at the Stahl- und Walzwerk Marienhütte GmbH in Graz.

To ensure the high quality standards, frequent ‘yield strength’ measurements are performed. These measurements have resulted in a large dataset which can now be analyzed and used to learn a prediction model.  First tests were promising and the thesis will be very likely a big success. 

The goal of this thesis is to analyze the data and to learn prediction models taking uncertainty estimates into account.

The models will be implemented and tested in Python.

Tentative Work Plan

To achieve our aim, the following concrete tasks will be focused on:

• Literature research on the underlying physical & chemical processes.
• Data analysis, filtering, preprocessing, visualization of the existing data. 
• Implementation of  deep neural networks (Variational Autoencoder), neural processes and GPs in Python. Baseline implementations are existing.
• Visualization and analysis of the prediction performance. An outlier detection and warning system should be implemented.
• (Optional) Implementation of neural time-series models like LSTMs.
• Analysis and evaluation of the provided data.

Supervisors: Honghu Xue, Elmar Rückert

Finished: 22.April 2022

Abstract

In this thesis, we propose an approach that combines reinforcement learning and automatic curriculum learning to solve a visual navigation task. A pedestrian agent is expected to learn a policy from scratch in a street-crossing scenario in a realistic traffic simulator CARLA. For this, the pedestrian is restricted to its first-person perspective as sensory input. The pedestrian cannot obtain full knowledge of the environment, which raises a partial observability challenge. To achieve this, an improved version of the Distributional Soft Actor-Critic algorithm is implemented. The algorithm adopts a newly proposed 3D dilated convolutional architecture to deal with the partial observability problem. To further improve its performance, we develop an automatic curriculum learning algorithm called NavACL+ on top of NavACL. As suggested in the results and ablation studies, our approach outperforms the original NavACL by 23.1%. Additionally, the convergence speed of NavACL+ is also observed to be 37.5% quicker. Moverover, the validation results show that the trained policies of NavACL+ are much more generalizable and robust than other variants in terms of different initial starting poses. NavACL+ policies perform 28.3% better than other policies training from a fixed start.

Thesis

Solving Visual Navigation Tasks for Pedestrian Trajectory Generation Using Distributional Reinforcement Learning and Automatic Curriculum Learning in CARLA