Staged Development of Robot Skills: Behavior Formation, Affordance Learning and Imitation with Motionese

Abstract

Inspired by infant development, we propose a three staged developmental framework for an anthropomorphic robot manipulator. In the first stage, the robot is initialized with a basic reach-and-enclose-on-contact movement capability, and discovers a set of behavior primitives by exploring its movement parameter space. In the next stage, the robot exercises the discovered behaviors on different objects, and learns the caused effects; effectively building a library of affordances and associated predictors. Finally, in the third stage, the learned structures and predictors are used to bootstrap complex imitation and action learning with the help of a cooperative tutor. The main contribution of this paper is the realization of an integrated developmental system where the structures emerging from the sensorimotor experience of an interacting real robot are used as the sole building blocks of the subsequent stages that generate increasingly more complex cognitive capabilities. The proposed framework includes a number of common features with infant sensorimotor development. Furthermore, the findings obtained from the self-exploration and motionese guided human-robot interaction experiments allow us to reason about the underlying mechanisms of simple-to-complex sensorimotor skill progression in human infants.

Language Meddles with Infants’ Processing of Observed Actions

“…It becomes then relevant to understand in depth how the different processes involved in tutoring and learning interact. Drawing inspiration from the current human developmental study, we propose that future robots should start their learning by anticipating tutors' goal through action observation (Butz et al, 2003;Theofilis et al, 2013;Ugur et al, 2015). However this process should be overridable by any linguistic signal simultaneously provided by the tutor, averting robot's attention from the action target and potentially directing it toward how a goal is achieved.…”

“…Any source of information, be it the observation of their parents' actions or their verbal instruction, are exploited to rapidly acquire new action and linguistic competencies. Enabling robots to learn to understand human actions as human infants do is a promising research line pursued in current robotics research (e.g., Rohlfing et al, 2006;Cangelosi et al, 2010;Ugur et al, 2015). However, it is not yet clear how infants integrate efficiently the multimodal tutoring inputs provided simultaneously by their parents, in order to understand their actions.…”

Bottom-up learning of object categories, action effects and logical rules: From continuous manipulative exploration to symbolic planning

“…Previously we showed that a robot that is initialized with a basic reach-and-grasp movement capability can discover a set of action primitives, learn a library of affordances and associated predictors, and finally use these structures to bootstrap complex imitation with the help of a cooperative tutor [10]. The methods provided in the current study enable the robot to reach a higher level of cognitive competence, where it forms structures for planning at a symbolic level.…”

Discovering and Manipulating Affordances

“…The approach was based on a cycle of exploration-manipulation, initialized with a collection of minimal knowledge and innate capabilities. These works studied the discovery of meaningful discrete motion primitives [10] or sequences thereof [8], using stochastic and deterministic [16] approaches. This allowed a robot to learn object affordances and the predictors that anticipate the effects that these action primitives create.…”