IMPACT - Intrinsically Motivated Planning Architecture for Curiosity-driven roboTs
It is a research collaboration between the Planning and Scheduling Technology Laboratory (PST) and the Laboratory of Computational Embodied Neuroscience (LOCEN). This activity aims to study and develop prototypes of software architectures able to control robotic systems as well as complex subsystems operating in space. This problem will be addressed through an interdisciplinary approach integrating techniques of Artificial Intelligence (IA), Autonomous Robotics and Machine Learning. In particular, state-of-the-art AI planning systems and algorithms will be integrated, with reinforcement learning algorithms guided by intrinsic motivations (curiosity, exploration, novelty, surprise). The aim of the research is to: (i) develop a software system that allows a robotic platform to represent the skills autonomously learned through intrinsic motivations during the exploration of the environment, in an appropriate symbolic form; (ii) use the symbolic planning ability to improve the autonomous acquisition of possible additional skills.