 Robotic Systems LabOpen OpportunitiesMinimal is a mostly 3D-printed, highly reconfigurable robot. Using state-of-the-art reinforcement learning, we will explore novel and highly advanced hardware design possibilities that will be coupled with design optimization through learning. This will enable the next generation of robots to be a lot faster, stronger and agile. - Engineering and Technology
- Master Thesis, Semester Project
| We want to develop a generalist digging agent that is able to do multiple tasks, such as digging and moving loose soil, and/or control multiple excavators. We plan to use decision transformers, trained on offline data, to accomplish these tasks. - Intelligent Robotics
- Master Thesis, Semester Project
| Searching for specific objects within a confined space requires advanced spatial and perceptual reasoning. In this project, we want to develop a learning-based system that finds a desired object in a cluttered scene efficiently and safely. - Computer Vision, Knowledge Representation and Machine Learning, Learning, Memory, Cognition and Language
- Master Thesis
| Model-based state estimation for locomotion has shown some significant drawbacks, especially in the case of complex contact scenarios. At the same time, locomotion controllers are evolving, now purposely using knee contacts or wheel slippage for advanced motions. The current model-based state estimation techniques often cannot supply sufficiently accurate observations for these controllers, leading to major estimation drifts and thus potential failures. In this project, we aim to leverage learning-based methods not only for locomotion control, but also for state estimation. Preliminary work shows that creating a state estimation through supervised learning from recorded simulation data can produce a viable solution. Furthermore, fusing these approaches with classical filtering theory opens a promising realm of research. The project should also compare the developed methods with existing approaches on real hardware. If time permits, we are interested in learning state estimation and locomotion jointly. - Intelligent Robotics, Robotics and Mechatronics
- Master Thesis, Semester Project
| Robots, like humans, should be able to use different parts of their morphology (base, elbow, hips, feet) for interaction. This project focuses on learning multi-modal interactions from demonstrations for mobile manipulators. - Intelligent Robotics, Knowledge Representation and Machine Learning
- Master Thesis, Semester Project
| This project addresses the task of 6D pose estimation for general-purpose objects, particularly when dealing with occlusion. We aim to leverage recent deep learning methods and synthetic data generation schemes to enable robust object manipulation. - Intelligent Robotics
- Master Thesis, Semester Project
| This project explores wheeled-legged legged robots, i.e., a robot that has both wheels and point-feet as end-effectors of its legs. Thereby, different locomotion modes should be explored, as well as different configurations of mounting wheels to legs. One idea could be a diagonal bicycle mode, another could be optimizing locomotion for payload transport. The project should include the implementation and deployment of the developed locomotion concepts and policies on real hardware. - Intelligent Robotics, Robotics and Mechatronics
- Master Thesis, Semester Project
|
Transport of packages of various dimensions is often mentioned as one of the most viable use cases for autonomous mobile robots. The ability to autonomously pick up and self-load a package is, however, a functionality that many systems are still lacking. Preliminary work showed that quadrupedal robots have the potential to execute this skill by manipulating payloads with their legs or main body. In this project, we aim to investigate how legged and wheeled legged robots can achieve autonomous package pick-and-load tasks with practical design modifications and clever maneuvers.
- Intelligent Robotics, Mechanical Engineering, Robotics and Mechatronics
- Master Thesis, Semester Project
| Legged robots rely on the assessment of geometric and visual data to determine where they can navigate safely. While today's learning-based methods show great performance in the assessment of the environment, we cannot guarantee that all hazards can be identified correctly and, as a result, cannot generalize to arbitrary environments. Consequently, we have to build resilient navigation algorithms that can take into account these failure cases and allow, by interaction with the environment (touching, bumping), to reactively update the planned path. In our previous work [1], we showcased the potential for using end-to-end reinforcement learning to tackle this challenge, which builds upon [4,5]. - Computer Perception, Memory and Attention, Intelligent Robotics, Knowledge Representation and Machine Learning
- Master Thesis, Semester Project
| Achieving task-level generalizations requires acquiring a large amount of rich interaction data. Simulators offer a safe, efficient, and cost-effective means of system development. However, generating task and scene-level diversity requires significant human effort to develop and verify novel tasks. This project aims to automate this process by leveraging the grounding capabilities of large language models. - Intelligent Robotics, Knowledge Representation and Machine Learning
- Bachelor Thesis, Master Thesis, Semester Project
|
|
