Band 3, Heft 4

Infants extend their repertoire of behaviours from initially simple behaviours with single objects to complex behaviours dealing with spatial relationships among objects. We are interested in the mechanisms underlying this development in order to achieve similar development in artificial systems. One mechanism is sensorimotor differentiation, which allows one behaviour to become altered in order to achieve a different result; the old behaviour is not forgotten, so differentiation increases the number of available behaviours. Differentiation requires the learning of both sensory abstractions and motor programs for the new behaviour; here we focus only on the sensory aspect: learning to recognise situations in which the new behaviour succeeds. We experimented with learning these situations in a realistic physical simulation of a robotic manipulator interacting with various objects, where the sensor space includes the robot arm position data and a Kinect-based vision system. The mechanism for learning sensory abstractions for a new behaviour is a component in the larger enterprise of building systems which emulate the mechanisms of infant development.

The use of concepts is fundamental to human-level cognition, but there remain a number of open questions as to the structures supporting this competence. Specifically, it has been shown that humans use concept prototypes, a flexible means of representing concepts such that it can be used both for categorisation and for similarity judgements. In the context of autonomous robotic agents, the processes by which such concept functionality could be acquired would be particularly useful, enabling flexible knowledge representation and application. This paper seeks to explore this issue of autonomous concept acquisition. By applying a set of structural and operational principles, that support a wide range of cognitive competencies, within a developmental framework, the intention is to explicitly embed the development of concepts into a wider framework of cognitive processing. Comparison with a benchmark concept modelling system shows that the proposed approach can account for a number of features, namely concept-based classification, and its extension to prototype-like functionality.

Robots and living beings exhibit latencies in their sensorimotor processing due to mechanical and electronic or neural processing delays. A reaction typically occurs to input stimuli of the past. This is critical not only when the environment changes (e.g. moving objects) but also when the agent itself moves. An agent that does not predict while moving may need to remain static between sensory input acquisition and output response to guarantee that the response is appropriate to the percept. We propose a biologically-inspired learning model of predictive sensorimotor integration to compensate for this latency. In this model, an Elman network is developed for sensory prediction and sensory filtering; a Continuous Actor-Critic Learning Automaton (CACLA) is trained for continuous action generation. For a robot docking experiment, this architecture improves the smoothness of the robot’s sensory input and therefore results in a faster and more accurate continuous approach behavior.

Robot-assisted therapy (RAT) is an emerging field that has already seen some success and is likely to develop in the future. One particular application area is within therapies for autism spectrum disorders, in which the viability of the approach has been demonstrated. The present paper is a vision paper with the aim of identifying research directions in the near future of RAT. Specifically, we argue that the next step in such therapeutic scenarios is the development of more substantial levels of autonomy which would allow the robot to adapt to the individual needs of children over longer periods of time (while remaining under the ultimate supervision of a therapist). We argue that this requires new advances on the level of robot controllers as well as the ability to infer and classify intentions, goals and emotional states of the robot’s interactants. We show that the state of the art in a number of relevant disciplines is now at the point at which such an endeavour can be approached in earnest.

The implementation of sequence learning in robotic platforms offers several challenges. Deciding when to stop one action and continue to the next requires a balance between stability of sensory information and, of course, the knowledge about what action is required next. The work presented here proposes a starting point for the successful execution and learning of dynamic sequences. Making use of the NAO humanoid platform we propose a mathematical model based on dynamic field theory and reinforcement learning methods for obtaining and performing a sequence of elementary motor behaviors. Results from the comparison of two reinforcement learning methods applied to sequence generation, for both simulation and implementation, are provided.