Sustaining learning in critical domains of robotic systems
Hoimonti Pal · Journal of computing sciences in colleges · 2002
Previous research in the field of Artificial Intelligence has focused on systems that are based on basic ideas of human long-term learning. These systems are not easily applicable to critical situations for robotic systems. In risky situations robotic systems use hierarchical/instinctual memory to address these perilous situations. However, this instinctual memory is often modified over long periods of time resulting in a sustained long-term memory system that interacts with the hierarchical system. In this paper a new method to modify and create learning in hierarchical/instinctualmemory is introduced in order to create a robotic system that is more capable of surviving in critical states. This new procedure draws on Learning Theory from the field of Cognitive Psychology in order to create and maintain this learning system for robotic structures. The new system that was created was first developed by re-analyzing the basic structure of memory and learning systems previously used in robotic systems. It is composed of six basic structures-Instinctual learning, Short-term learning, Long-term learning, motor memory, short-term memory, and long-term memory/knowledge base. Motor memory closely interacts exclusively with Instinctual Learning in a bi-directional relationship. Instinctual Learning also interacts with itself. The short term learning system interacts with both short-term memory as well as short-term learning itself. Finally, the long term learning system interacts with both long-term memory/knowledge base and long term learning. Both long-term learning and long-term memory can modify instinctuallearning; instinctuallearning can also modify long-term learning. The short and long- term learning structures interact and can modify one another. All modifications take place due to the repetitiveness and/or distinctiveness of the actions specified. For example, instinctual learning is self-modifiable through repetition in certain states, and this will also modify long-term learning and the long-term knowledge base. This system creates structures to deal with particular situations, but these distinct structures are also highly interactive. This provides a means to create sustained learning in risky domains. This system will allow practical robotic systems in critical domains to learn faster. This scheme has breakthrough potential for practical robotic systems such as military defense, police and rescue situations, medical diagnostic systems and other critical domain applications.