Integrating Open-Ended Learning in the Sense-Plan-Act Robot Control Paradigm

This paper presents the achievements obtained from astudy performed within the IMPACT (Intrinsically Motivated Plan-ning Architecture for Curiosity-driven roboTs) Project funded by theEuropean Space Agency (ESA). The main contribution of the workis the realization of an innovative robotic architecture in which thewell-knownthree-layeredarchitectural paradigm (decisional, execu-tive, and functional) for controlling robotic systems is enhanced withautonomous learningcapabilities. The architecture is the outcomeof the application of an interdisciplinary approach integrating Artifi-cial Intelligence (AI), Autonomous Robotics, and Machine Learning(ML) techniques. In particular, state-of-the-art AIplanning systemsand algorithms were integrated with Reinforcement Learning (RL)algorithms guided byintrinsic motivations(curiosity, exploration,novelty, and surprise). The aim of this integration was to: (i) developa software system that allows a robotic platform to autonomouslyrepresent in symbolic form theskillsautonomously learned throughintrinsic motivations; (ii) show that the symbolic representation canbe profitably used for automated planning purposes, thus improvingthe robot's exploration and knowledge acquisition capabilities. Theproposed solution is validated in a test scenario inspired by a typicalspace exploration mission involving a rover.