The SENSOPAC project will combine machine learning techniques and modelling of biological systems to develop a machine capable of abstracting cognitive notions from sensorimotor relationships during interactions with its environment, and of generalising this knowledge to novel situations. Through active sensing and exploratory actions the machine will discover the sensorimotor relationships and consequently learn the intrinsic structure of its interactions with the world and unravel predictive and causal relationships. Together with action policy formulation and decision making, this will underlie the machine’s abilities to create abstractions, to suggest and test hypotheses, and develop self-awareness. The project will demonstrate how a naïve system can bootstrap its cognitive development by constructing generalization and discovering abstractions with which it can conceptualize its environment and its own self. The continuous developmental approach will combine self-supervised and reinforcement learning with motivational drives to form a truly autonomous artificial system. Throughout the project, continuous interactions between experimentalists, theoreticians, engineers and roboticists will take place in order to coordinate the most rigorous development and testing of a complete artificial cognitive system.
Site – http://www.sensopac.org