Max Planck ETH Center for Learning Systems

Open Opportunities

Securing the Future: Enhance Human Safety in Robot Decision Making ETH Zurich Robotic Systems Lab Robots have become increasingly advanced recently, capable of performing challenging tasks such as taking elevators and cooking shrimp. Moreover, their ability to accomplish long-horizon tasks given simple natural language instructions is also made possible by large language models. However, with this increased functionality comes the risk that intelligent robots might unintentionally or intentionally harm people based on instructions from an operator. On the other hand, significant efforts have been made to restrain large language models from generating harmful content. Can these efforts be applied to robotics to ensure safe interactions between robots and humans, even as robots become more capable? This project aims to answer this question. Engineering and Technology, Information, Computing and Communication Sciences Master Thesis, Semester Project

Vison- and Depth-based Reinforcement Learning for dexterous and accurate manipulation with Aerial Robots [multiple openings] ETH Zurich Autonomous Systems Lab Our team develops novel Aerial Robots that are able to autonomously manipulate and perform work in flight. In this thesis, we would like to explore the learning of task-specific policies for manipulation in flight. Intelligent Robotics Master Thesis

Error bounds for scalable Gaussian process regression ETH Zurich Research Zeilinger The goal of the project consists in deriving error bounds for the approximate Gaussian process regression method given by the FITC method. Engineering and Technology Master Thesis, Semester Project

Error bounds for Regularized Trigonometric Regression in the Multi-task setting ETH Zurich Research Zeilinger Multi-task learning is the problem of jointly learning multiple functions that are “related” to each other. By leveraging this similarity, estimation performance can be improved on each (possibly unseen) task, and one can make an efficient use of the available data. The project aims at deriving uncertainty bounds around the multi-task-system estimates. Specifically, the candidate will work with the regularized trigonometric regression inspired by the so-called sparse-spectrum Gaussian process regression, investigate the issue of bias learning (i.e., finding the features that encode similarity among tasks) and derive error bounds for it, possibly setting the analysis in the statistical learning framework. Engineering and Technology Master Thesis, Semester Project

Learning-based stochastic Model Predictive Control with scalable Gaussian process regression ETH Zurich Research Zeilinger One of the key ingredients in Model Predictive Control (MPC) schemes is an effective model of the dynamical system’s response to external inputs. However, first- principles models are often not accurate enough, as there might be unknown external disturbances and model mismatches. To address this, learning-based control aims at complementing nominal models with data-based ones, which can be refined online as new system observations are gathered. Thus, such a model should be both expressive and fast to update. This project focuses on a learning-based stochastic MPC scheme, where uncertainty in the model is learned with an approximate Gaussian process, namely the regularized trigonometric regression stemming from the so- called sparse-spectrum Gaussian processes. To this aim, the candidate will review the available uncertainty bounds around these approximate Gaussian-process-based estimates and incorporate them in the MPC formulation. The chance-constraints thereby obtained are then to be analyzed to rigorously prove recursive feasibility and stability of the closed-loop system. Engineering and Technology Master Thesis, Semester Project

Organ-on-Chip Automation Hardware Design: Development and Integration of a Miniature Robotic XY Stage for a Miniature Microscope ETH Zurich Sensory-Motor Systems Lab Organs-on-Chip (OoC) replicate human organs in vitro but often lack physiological perfusion profiles. This project aims to develop a compact robotic XY stage for integration with OoC perfusion systems, enhancing automation and precision. By improving compatibility and mimicking dynamic blood flow, this innovation advances research and pharmaceutical applications. Biomedical Engineering Master Thesis

Enhancing Bone Segmentation in Ultrasound Imaging Using Physics-Informed Deep Learning Models University of Zurich Research in Orthopedic Computer Science Other organizations: Computer Vision and Geometry Group Computer-Assisted Orthopedic Surgery (CAOS) has been demonstrated to improve surgical precision in various procedures, including spinal fusion surgery, arthroplasty, and bone deformity correction [1,2]. Ultrasound, as a radiation-free, cost-effective, and portable alternative to CT and X-ray imaging, has been employed for real-time visualization of both soft tissues and bones through the reflection of acoustic waves. Despite its advantages, ultrasound imaging has inherent limitations such as low signal-to-noise ratio, acoustic shadowing, and speckle noise, which pose challenges for interpretation by surgeons. In our project, we have collected a dataset comprising over 100k ultrasound images with precise bone annotations. These bone labels are categorized into two classes: high-intensity regions (high signal-to-noise ratio) and low-intensity regions (low signal-to-noise ratio), as shown in Figure 1. According to experiment results, surgeons’ performance for bone labeling for low-intensity regions declined significantly compared to the high-intensity regions. [1] Pandey, Prashant U., et al. "Ultrasound bone segmentation: A scoping review of techniques and validation practices." Ultrasound in Medicine & Biology 46.4 (2020): 921-935. [2] Hohlmann, Benjamin, Peter Broessner, and Klaus Radermacher. "Ultrasound-based 3D bone modelling in computer assisted orthopedic surgery–a review and future challenges." Computer Assisted Surgery 29.1 (2024): 2276055. Computer Vision, Image Processing, Medical and Health Sciences, Pattern Recognition Bachelor Thesis, Master Thesis, Semester Project

Diffusion-based Shared Autonomy System for Telemanipulation ETH Zurich Robotic Systems Lab Robots may not be able to complete tasks fully autonomously in unstructured or unseen environments, however direct teleoperation from human operators may also be challenging due to the difficulty of providing full situational awareness to the operator as well as degradation in communication leading to the loss of control authority. This motivates the use of shared autonomy for assisting the operator thereby enhancing the performance during the task. In this project, we aim to develop a shared autonomy framework for teleoperation of manipulator arms, to assist non-expert users or in the presence of degraded communication. Imitation learning, such as diffusion models, have emerged as a popular and scalable approach for learning manipulation tasks [1, 2]. Additionally, recent works have combined this with partial diffusion to enable shared autonomy [3]. However, the tasks were restricted to simple 2D domains. In this project, we wish to extend previous work in the lab using diffusion-based imitation learning, to enable shared autonomy for non-expert users to complete unseen tasks or in degraded communication environments. Intelligent Robotics, Robotics and Mechatronics ETH Zurich (ETHZ), Semester Project

Instruction Space Characterization for Interactive Human-Robot Domains ETH Zurich Sensory-Motor Systems Lab Human intention drives goal-directed actions, requiring decomposition into hierarchical instructions, from abstract strategies to concrete interactions. Robotics seeks to automate these instructions, but challenges arise in domains like neurorehabilitation, where complex human-robot interactions introduce inefficiencies from incomplete automation, diverse user preferences for strategies, and difficulties to engage with technical parameters. This thesis will explore shared instruction space characteristics across robotic domains, focusing on bottlenecks in automating high-level instructions. Novel metrics will be developed that define and classify instruction layers, strategies, and requirements. Intelligent Robotics Bachelor Thesis

High dose high intensity proprioception training for neurological patients using a robotic device ETH Zurich Rehabilitation Engineering Lab Proprioception, often called the "sixth sense," is vital for coordinating movements and maintaining balance, especially in the hands and fingers. In neurological patients, impaired proprioception in the upper limbs can hinder daily tasks and reduce quality of life. Traditional rehabilitation often lacks the intensity and precision needed for optimal recovery of fine motor skills. This project leverages high-dosage training with the ETH MIKE, a validated one-degree-of-freedom robotic device designed for precise, repetitive movements to enhance proprioception and motor function. By promoting neuroplasticity and functional recovery, this research aims to advance rehabilitation practices. Participants will gain hands-on clinical experience, train with neurological patients, and collaborate with therapists, engineers, and researchers. Biomedical Engineering, Clinical Sciences, Human Movement and Sports Science, Medicine-general, Neurosciences, Other Bachelor Thesis, Internship, Master Thesis, Semester Project