The improvement of robotic grasping in the last years has mainly been pushed by new learning techniques. A promising direction is learning from demonstration, where the human operator teaches a robot the tricks of the specific task. On the other hand with these learning techniques it is hard to understand the underlying reasoning.
We plan on using augmented reality and virtual reality setups to teach and validate real world robotics grasping approaches.
This project aims at building a system that allows to train and validate robotic grasping.
There is different parts and scope possible for honors, research and PhD projects.
in particular, we want AR tracking of the world to visualise what the robot would do in the current situation. Human interaction with the scene should provide a clue for the to be built learning system.
Various scopes are possible, for example:
- Development of (at least a sub) system based on AR and simulation and current grasp the techniques.
- Implementation of a teaching system
- creating a deep learning system to improve grasping
- Integrate with a real robot
Contact the supervisor for more information.