How Hard Does a Robot Hit? Learning 3D Collision Mass Maps for Safe Human-Robot Interaction

Institute

Professur für Cyber Physical Systems (TUM-CIT)

Type

Bachelor's Thesis / Semester Thesis / Master's Thesis /

Content

experimental / constructive /  

Description

You will build a 3D, posture-aware CMM for real 7-DoF robots by combining experiments, data processing, and machine learning:
• Data: Run automated collision experiments on different 7-DoF robots and build a pipeline that turns raw force measurements into a clean effective mass dataset.
• Learning: Predict the effective mass from collision location, direction, and posture by learning a correction to the physics-based model, with uncertainty estimates (e.g., Gaussian processes or neural networks) that keep the map on the safe side. 
• Active learning: Let the model choose where to measure next (e.g., Bayesian optimization) to minimize the number of collisions, with pre-training in simulation.
• Application: Use the map to select the safest posture and the fastest safe velocity along a path on the real robot.

Requirements

1. Programming in C++

2. Basic robotics and robot kinematics knowledge, an excellent grade in robotics-related courses is required.

3. Self-motivated

4. Interest in machine learning.

5. Regular on-site presence, since the collision experiments are carried out on real robots.

Possible start

01.11.2026

Contact

Yue Zhang
yue22.zhangtum.de

Announcement