Human-Inspired Robot Memory

Human-Inspired Robot Memory

Robots are expected to operate in real, human created environments and perform tasks designed by or for humans. Such environments cause a deluge of information which has to be adequately processed and stored. For example, if humans remember an episode of their life, they do not recall an exact video stream. Rather a subjective reconstruction of this episode influenced by other experiences is created. However, although human memories are incomplete and possibly incorrect they enable us to live (and survive) in a very complex world. So, in order to build a robot which operates in this world and is able to life-long learning - as we are - a memory system is required, which supports adequate information processing functions, e.g., selection of relevant information, abstraction, compression, association of new and already known knowledge, integration and forgetting. Since humans are able to achieve this easily, we assume that human-inspired information processing mechanisms, in particular human-inspired memory systems, are beneficial for robots as well.

Our research is aimed at developing approaches for equipping robots with technical memory systems that allow them to successfully learn and perform real-world tasks. Similar to humans, these systems are distinguished according to temporal (e.g., short-term and long-term memory) and functional aspects (e.g., working memory, semantic memory and episodic memory). However, robots cannot rely on such a massively parallel information processing organ as the human brain. They are rather limited to current computer hardware, which works more or less sequentially and is significantly less powerful. On the other hand, microprocessors are able to perform considerably more accurate computations than biological neurons. These differences have to be taken into account regarding the development of robot memories. Furthermore, knowledge from different fields of research is required, in particular, psychology, biology and informatics. So our research is of clearly inter-disciplinary character.

A crucial issue when designing robot systems is their architecture. It determines the way different software components of a robot are able to communicate. Therefore it has a significant impact on any approach to build a robot memory.  On the other hand, memory is a common property of numerous software modules. So it determines the architecture as well. Here, we investigate how the concept of a human-inspired memory can be incorporated in robot architectures in order to enhance the robots' capabilities. Are there general design principles? What are the advantages and drawbacks of such an approach?

Ausgeschriebene Bachelorarbeiten

• Online-Lernen von Bild-zu-Bild Assoziationen (26.8.2009)