Action and Adaptation: Lessons from Neurobiology and Challenges for Robot Cognitive Architectures

摘要

In this paper we review recent findings from neurobiology concerning action and adaptation, from the level of motor control up to the level of decision making. At the level of motor control, proposals concerning multiple forward and inverse internal models in the cerebellum are described. Particular consideration is devoted to the adaptive mechanisms of these models. Then, the decision making processes taking place in the basal ganglia and the orbitofrontal cortex are discussed. Under the light of recent findings, these two structures correspond to two levels of decision making with distinct characteristics. The amygdala plays an important role in these decision making processes, for it provides affective evaluation to the action options being considered, as well as an adaptive role in associating novel stimuli with affective states. Then, we discuss possible contributions that these findings may provide for the development of cognitive architectures for robots. In particular, we focus on the following issues: (1) the integration of information coming from different levels, with distinct natures and time frames (the binding problem), and (2) the nature of the described internal models, in the sense that they model the relationship of an embedded and embodied being with its environment. We finally discuss future directions for the research on cognitive architectures, taking into account the discussed neurobiological adaptive mechanisms.