From learning to new goal generation in a bioinspired robotic setup

In the field of cognitive bioinspired robotics, we focus on autonomous development, and propose a possible model to explain how humans generate and pursue new goals that are not strictly dictated by survival. Autonomous lifelong learning is an important ability for robots to make them able to acquire new skills, and autonomous goal generation is a basic mechanism for that. The Intentional Distributed Robotic Architecture (IDRA) here presented intends to allow the autonomous development of new goals in situated agents starting from some simple hard-coded instincts. It addresses this capability through an imitation of the neural plasticity, the property of the cerebral cortex supporting learning. Three main brain areas are involved in goal generation, cerebral cortex, thalamus, and amygdala; these are mimicked at a functional level by the modules of our computational model, namely Deliberative, Working-Memory, Goal-Generator, and Instincts Modules, all connected in a network. IDRA has been designed to be robot independent; we have used it in simulation and on the real Aldebaran NAO humanoid robot. The reported experiments explore how basic capabilities, as active sensing, are obtained by the architecture.