Master Projects

The Chair of Digital Health with a Focus on Data Science is looking for talented Master’s students to undertake our available master thesis projects (listed below).

1) Advancing Human-Robot Collaboration for Rehabilitation: Online Reinforcement Learning for Variable Admittance Control

The research objective is to develop an online reinforcement learning (OnRL) approach tailored to variable admittance control in the context of human-robot collaboration. This research aims to create a robust OnRL framework that optimizes robot control policies through simulation-based learning, specifically focusing on variable impedance control. The primary goal is to improve the adaptability and safety of robots when collaborating with humans. This will be achieved by enhancing the robot's ability to adjust its stiffness and compliance dynamically based on the task and the interaction with human partners. The research will also address safety concerns by integrating safety mechanisms into the RL framework, enabling the robot to predict, react to, and mitigate potential risks during collaborative tasks.

Prerequisites: Coding language (python preferable), Machine Learning (basics), Reinforcement Learning (preferable)

Supervisor: Prof. Dr. Aldo A. Faisal

Co-supervisor: Dr. Jyotindra Narayan

2) Offline Reinforcement Learning for Human-Like Gait Control in Neurological Patients

This research plan is to focus on investigating the feasibility of employing an offline reinforcement learning (OffRL) approach for neuromuscular gait modeling, which could be particularly beneficial in rehabilitation. By optimizing the reward policies, we will try to train the model to establish sensory-motor mappings (control policy), enabling it to generate human-like walking patterns. The training process incorporated essential factors such as human motion capture data, muscle activation patterns, and metabolic cost estimation within the reward function. Our goal is to demonstrate the model's ability to faithfully reproduce human kinematics and ground reaction forces during walking and generate human-like walking behavior at different speeds, with a focus on improving walking movements in neurological patients during rehabilitation.

Prerequisites: Coding language (python preferable), Machine Learning (basics), Reinforcement Learning (preferable)

Supervisor: Prof. Dr. Aldo A. Faisal

Co-supervisor: Dr. Jyotindra Narayan

3) Human interfacing via surface electromyography with biofeedback of custom features

Advancements in recording methods for neurophysiological signals have facilitated the study of the human motor system with increasing accuracy and number of channels. In particular, the use of high-density (>32 channels) surface electromyography (EMG) results in increased spatial resolution but also opens the possibility of decoding the neural drive to muscles. In this context, the higher number of electrode channels allows the detection of individual motoneuron action potentials and spike trains through the technique known as motor unit decomposition. Moreover, the addition of biofeedback related to these signals (e.g. by showing them in a user-friendly visual interface) combined with consistent training can help users attain improved control over their nerve signals. Hence, this project aims to develop a platform that aids humans to be better at controlling devices using EMG-based interfaces. The project entails developing and validating a visual interface that can show in real-time signals acquired through high-density EMG and provides biofeedback on, for instance, the decomposed motoneuron spike trains, or other additional features.

Prerequisites: Matlab and/or Python, basic statistics, signal processing (preferable)

Supervisor: Prof. Dr. Aldo A. Faisal

4) Interface modalities comparison to control a supernumerary robotic finger

The attachment and interfacing of artificial robotic limbs to the human body have been extensively studied in the field of prosthetics. On the other side, the idea of attaching these robotic limbs to able-bodied people for practical applications started being studied formally only over the last decade, although examples can be found since the year 1980. These devices are considered by several researchers as the next step in human evolution given that supernumerary (i.e. in addition to the normal number) limbs, by adding more degrees of freedom to the human body, enable the wearer to perform novel tasks that could not be achieved without them. Because of their human-enhancing features, supernumerary robotic limbs (SRL) are considered augmentation devices. We previously developed and tested an extra robotic finger system, the Supernumerary Robotic 3rd Thumb (SR3T), on a piano-playing task. Now, we aim to analyze the effect that different interfacing modalities can have on the performance when using an extra limb. The project entails adapting current existing hardware (the SR3T) to be controlled using different modalities (e.g. inertial sensors, EMG) in a motor task, such as key pressing. More on the extra robotic thumb can be found here.dvancements in recording methods for neurophysiological signals have facilitated the study of the human motor system with increasing accuracy and number of channels. In particular, the use of high-density (>32 channels) surface electromyography (EMG) results in increased spatial resolution but also opens the possibility of decoding the neural drive to muscles. In this context, the higher number of electrode channels allows the detection of individual motoneuron action potentials and spike trains through the technique known as motor unit decomposition. Moreover, the addition of biofeedback related to these signals (e.g. by showing them in a user-friendly visual interface) combined with consistent training can help users attain improved control over their nerve signals. Hence, this project aims to develop a platform that aids humans to be better at controlling devices using EMG-based interfaces. The project entails developing and validating a visual interface that can show in real-time signals acquired through high-density EMG and provides biofeedback on, for instance, the decomposed motoneuron spike trains, or other additional features.

Prerequisites: Matlab and/or Python, basic statistics, signal processing (preferable)

Supervisor: Prof. Dr. Aldo A. Faisal

Are you interested?

If interested in projects 1 or 2, please contact Prof. Dr. Aldo Faisal (aldo.faisal@uni-bayreuth.de) and Dr. Jyotindra Narayan (jyotindra.narayan@uni-bayreuth.de).

If interested in projects 3 or 4, please contact Prof. Dr. Aldo Faisal (aldo.faisal@uni-bayreuth.de) and Renato Mio (renato.mio-zaldivar@uni-bayreuth.de). In your email, please state your interest and a short description of your previous knowledge/experience.