Skeleton-based Human Motion Prediction for Safe Human-Robot-Collaboration
Currently, there is an increasing trend towards human-centric automation since suitable hardware like collaborative robots (cobots) became available. Nevertheless, pro-active instead of re-active safety measures are still lacking. Previous research at the professorship developed a deep neural network, which predicts discrete classes of human movement patterns based on partially observed movements.
This line of research shall now be extended to enable a similar network to predict a complete human movement trajectory. A planned motion of the robot could then be checked for possible collisions and adjusted accordingly. After a network has been developed and trained, it should additionally be evaluated using an existing robot research platform.
Cameras including software for recording movements
and a capable GPU-workstation and the robot setup will be made available.
Advisor:
- Florian Röhrbein, florian.roehrbein@…
Requirements:
- Strong mathematical foundations
- Deep Learning fundamentals
- Basic knowledge of robotics