Adna is presenting slides on the updates. 
Train project extension planned for late 2024.
Adna will continue to work on the ATARi simulation implementation in 2022.
Ege and Adna are supervising a master student on the ATARI games.
Adan and Ege are working on an ethics proposal where a submission is planned for April.

Details 1: Adna did completed first experiments to train the RL algorithms in the ATARI games from human data. Humans generate tuples of state/action/reward sequences. Evaluated are the performance, the cognitive load and human feedback of how they perceived the learning behavior (satisfaction & trust, compare to own playing).
Open question on how to improve the training performance. Elmar suggested to analyse the effect of signalling the learning performance to the human instructor, e.g., compared to the score increase in the last 2 minutes, currently the progress is faster or slower.
Fotios also worked on an related project using soft actor-critic. He will share his code and results with Adna.

Details 2: Ege is working on the wire maze and stacking task. Cognitive load, trust, etc. are investigated. Shared autonomy is investigated in transfer learning scenarios of learning multiple related tasks. Shared autonomy might be also investigated in teleportation tasks. Ege ist working on the Franka HW/SW framework. 

Framework: Philipp sugested to extend the framework overview scheme contuniously. At least an outlook of open questions and directions of future reseach can be presented in a final report.