Can a robot's adaptive behavior be animal-like without a learning algorithm?

摘要

The purpose of this paper is to investigate what task and to what extent a robot loading CBA (Consciousness-based Architecture) can achieve without learning algorithm. CBA is developed by the author's group, a five-layered hierarchical architecture of the relationship between consciousness and behavior: the hierarchy has the evolutionary arrangement of behaviors from reactive behaviors to symbolic ones. But no leaning algorithm is explicitly embedded in CBA. Behavioral outputs with CBA simulating fish, rat and ape, were compared with the two typical psychological experiments, discrimination learning, stochastic learning, by Bitterman. Results of the simulation show that CBA up to lower mammal level is successful in equivalents to the learnings. CBA also explains that it can accept conditioning by reinforcement stimuli. CBA, however, needs rule-based knowledge/learning algorithm for stochastic learning at ape level.