Bachelor / Master thesis topics
This student project focuses on collaborative furniture assembly to promote ergonomic working conditions. The aim is for humans and robots to form a symbiotic team and communicate by means of visual and auditory signals. At the same time, safety is ensured by sensory monitoring. The versatile assembly robot should adapt ergonomically to the worker by holding and positioning the workpiece ergonomically. Three processes (assembly, disassembly and mechanical processing) and two levels of abstraction (Lego- Duplo and wooden structures with screw connections) are to be experimentally set up and demonstrated. The test stand is used for interaction research between humans and robots in the context of Industry 5.0.
Requirements: Basic Knowledge in robotics
Advisor: Sascha Kaden
Development of an assistance robot for the individual production of welding processes without time-consuming "teaching". The robot holds the component while the human carries out/suggests the welding process. This division enables the human to react to changes in the joining gap and produce precise welded joints. The flexible object positioning by the robot should enable safety through optimal welding and at the same time increase ergonomics and process efficiency. The aim of the experiment is to set up a possible welding process with its constraints and to test the interaction between humans and robots.
Requirements: Basic Knowledge in robotics
Advisor: Sascha Kaden
This research explores using spiking neural networks, inspired by the brain's communication methods, to directly control robots. This could lead to more efficient and adaptable robot behavior.
This topic delves into creating models based on the brain's decision-making processes to predict robot actions. This could allow robots to react and behave more intelligently in real-time.
Here, the focus is on developing algorithms that enable robots to learn and complete 3D shapes from partial data. This could be crucial for tasks like object recognition and manipulation.
This research explores how robots can improve obstacle avoidance using a combination of reinforcement learning and experience aggregation. Reinforcement learning allows the robot to learn from its own trial and error, receiving rewards for avoiding obstacles and penalties for collisions. Experience aggregation groups similar situations together, enabling the robot to learn from the experiences of other robots as well. This combined approach aims to develop robots that can learn autonomously and adapt to new environments, leading to more efficient obstacle avoidance.
This topic tackles implementing facial recognition technology on a low-power computer like Raspberry Pi. This could be useful for applications like robot security or social interaction.
This research investigates using 3D scene graphs to improve robot understanding of their surroundings. These graphs act like maps, showing not only how far objects are (metric) but also what kind of objects they are (semantic). This allows robots to navigate their environment more effectively.
This research investigates using 3D scene graphs to improve robot understanding of their surroundings. These graphs act like maps, showing not only how far objects are (metric) but also what kind of objects they are (semantic). This allows robots to navigate their environment more effectively.
This research aims to establish a standardized system for classifying and evaluating algorithms used in inverse deep reinforcement learning, where the desired outcome is known and the robot learns the necessary actions.
This research dives into developing algorithms for robots within the MIRO simulation environment . These algorithms will allow robots to estimate depth from images captured by MIRO's built-in stereo vision system. Depth perception is crucial for tasks like navigation and object manipulation. By analyzing the data from the stereo cameras, alongside other potentially relevant sensors like ultrasonic ranging and infrared cliff sensors, the algorithms will essentially give robots a sense of 3D space. This enhanced perception will allow them to plan movements, avoid collisions, and interact with objects more effectively.
This research tackles enabling robots to identify suitable grasping points for objects based on 3D point cloud data, which represents the object's surface using data points.
This research investigates using deep learning to tackle a key challenge in robotics: estimating the 3D pose (both position and orientation) of everyday objects. Accurate pose estimation is crucial for robots to perform tasks like grasping household objects. The deep learning model is trained on a vast amount of data depicting various objects in different poses. This allows the robot to understand how the object is positioned and oriented, enabling it to plan and execute grasping maneuvers more effectively for successful manipulation.
Here, the research focuses on developing algorithms that enable mobile robots to make adaptable decisions and react to changes in their environment in real-time
This topic explores designing safe and efficient workspaces where humans and robots can collaborate effectively.