This paper describes research work aimed at designing realistic reasoning techniques for humanoid robots provided with advanced skills. Robots operating in real-world environments are expected to exhibit very complex behaviors, such as manipulating everyday objects, moving in crowded environments or interacting with people, both socially and physically. Such yet to be achieved capabilities pose the problem of being able to reason upon hundreds or even thousands different objects, places and possible actions to carry out, each one relevant for achieving robot goals or motivations. This article proposes a functional representation of everyday objects, places and actions described in terms of such abstractions as affordances and capabilities. The main contribution is twofold: (i) affordances and capabilities are represented as neural maps grounded in proper metric spaces; (ii) the reasoning process is decomposed into two phases, namely problem awareness (which is the focus of this work) and action selection. Experiments in simulation show that large-scale reasoning problems can be easily managed in the proposed framework.
Problem awareness for skilled humanoid robots
MASTROGIOVANNI, FULVIO;SCALMATO, ANTONELLO;SGORBISSA, ANTONIO;ZACCARIA, RENATO UGO RAFFAELE
2011-01-01
Abstract
This paper describes research work aimed at designing realistic reasoning techniques for humanoid robots provided with advanced skills. Robots operating in real-world environments are expected to exhibit very complex behaviors, such as manipulating everyday objects, moving in crowded environments or interacting with people, both socially and physically. Such yet to be achieved capabilities pose the problem of being able to reason upon hundreds or even thousands different objects, places and possible actions to carry out, each one relevant for achieving robot goals or motivations. This article proposes a functional representation of everyday objects, places and actions described in terms of such abstractions as affordances and capabilities. The main contribution is twofold: (i) affordances and capabilities are represented as neural maps grounded in proper metric spaces; (ii) the reasoning process is decomposed into two phases, namely problem awareness (which is the focus of this work) and action selection. Experiments in simulation show that large-scale reasoning problems can be easily managed in the proposed framework.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.