Developmental reasoning and planning with robot through enactive interaction with human

Overview & Synopsis

From automata to robots, the Human has always been fascinated by machines which could execute tasks for him, in several domains like industry or services. Indeed, we have used a developmental approach, where the robot has to learn new tasks during his life. Inspired by theories in child development, we have extracted the interesting concepts to implement them on a humanoid robotic platform : the iCub. Language acquisition is a first step, where the capacity to classify closed and opened class words allows to obtain a syntax which help the children to make the link between a sentence and its meaning. This method has been implemented with a recurrent neural network, using a database provided from the human by interaction with the robot. The control of the language allows the children to participate in more complex actions, in particular cooperative tasks, where speech is required to negotiate the learning mode within several modalities. Implemented on the iCub and the Nao, this allows a real-time learning and to realize a shared plan. Eventually, we have studied the functioning of the autobiographical memory, crucial toremember episodes of his life, to extract predictions from and to apply them in the future. By recreating this memory in SQL, and by formatting the data in PDDL, the iCub is then capable of reasoning in function of his own experience, allowing him to solve the Tower of Hanoi problem without knowing the solution before