Student Thesis: Robot Learning from Demonstration - Designing the Data Collection Hardware

25.09.2026, Abschlussarbeiten, Bachelor- und Masterarbeiten

Redesign and build a demonstration device for robot learning from demonstration, taking it from an existing CAD model to a working prototype using CAD and 3D printing. Depending on progress, the device will be used to collect demonstrations and train robot learning models such as VLAs.

Motivation
Learning from demonstration (LfD) is a widely used approach for teaching robots new manipulation skills. An expert demonstrates a task and the robot learns a generalizable policy from a set of demonstrations. Recent vision-language-action (VLA) models show that such policies can generalize well, provided they are trained on enough high-quality demonstration data.

One central challenge of this approach is data collection. Teleoperation setups are expensive and usually bound to a specific robot and location, which limits how much data can be recorded in practice. Dedicated demonstration devices offer a compelling alternative. They allow humans to perform tasks in a natural way while recording data that can be transferred directly to the robot. The design of such a device has a strong influence on the quality of the recorded data and therefore on how well the robot can learn from it.


System
The setup consists of a robot arm equipped with a gripper, cameras and additional sensors (tactile, etc.). Smaller-scale hardware for collecting demonstrations already exists. A first design of a new demonstration device is available as a CAD model.


Research Project
In this project, you will take the new demonstration device from the existing CAD model to working hardware. This includes redesigning the CAD model based on experience with the current collection hardware and building functional prototypes with integrated sensors using 3D printing. Depending on progress, you will use the device to collect demonstrations and train state-of-the-art robot learning models (e.g. VLAs) that allow the robot to perform the demonstrated tasks autonomously. The work combines elements of mechanical design, mechatronics, robotics and machine learning.


Goals
- Review the existing CAD design and the current collection hardware and derive requirements for the new device.
- Redesign the demonstration device and build a functional prototype using CAD and 3D printing, including sensor integration.
- Validate the new device by collecting demonstration data for representative manipulation tasks.
- Optionally, train learning-from-demonstration models on the collected data and evaluate how reliably the robot performs the tasks autonomously.


Prerequisites
- Interest in robotics and robot learning, and a mechatronics-oriented approach to problem solving.
- Solid experience with CAD and 3D printing, ideally in a robotics context.
- Excited to build and iterate on physical prototypes.
- Prior experience with any of the following is a plus: robotics projects, mechatronic design, sensor integration, ROS, deep learning frameworks.
- Available to start the topic soon.


If you are excited about the topic but don't check every box, feel free to reach out anyway!

Kontakt: valdrin.aslani@tum.de