Building episodic memories for interactive systems


Suitable for:
Master project


Interactive systems such as social robots or virtual agents need to adapt to their users and learn from past interaction experiences. A prerequisite for this is a memory of past interaction episodes, built up through the course of many interactions. While classical approaches to develop such a memory are often pre-structured, current Machine Learning methods allow to build dynamic low-dimensional representations (“embeddings”) that can be used for efficient storage, retrieval, or completion of memorized episodes. This project aims to explore and develop an approach to develop such a memory component in our existing “Social Cognitive Systems Architecture” that drives socially intelligent agents. You will have to investigate and develop a suitable approach and test it through repeated human-robot interactions.


  • Basic knowledge and interest in cognitive systems/Machine Learning
  • Mathematical and programming skills (Python)
  • Ability and willingness to work in an advanced software environment