The RobotDoC research will focus around 16 projects linked to the 16 Fellows' research activities. The projects will investigate the development of different sensorimotor and cognitive capabilities in humanoid robots such as iCub and ASIMO. The list of research projects, and associate lead partners, is provided below. Please contact the lead researcher if you are interested in any of these research projects.
- The emergence of symbol composition capabilitiesUniversity of Plymouth, UK (Angelo Cangelosi, Marek Rucinski)
New research will address the following questions: How can continuous behavioural patterns (e.g. action and perceptual categories) form discrete conceptual representations? How can such discrete representation become independent from their analogical representation and be handled in an abstract symbol manipulation system? How can such a symbol system be used for various symbol-based compositional tasks (e.g. language, number knowledge, reasoning). The model and experiment will focus on number symbols and numerical cognition.
- The grounding of abstract categories University of Plymouth, UK (Angelo Cangelosi, Francesca Stramandinoli)
How can robots use sensorimotor categories (e.g. concept of space, internal feelings/emotions) to indirectly ground abstract concepts (e.g. time, happiness). What kind of metaphorical mechanisms are used to combine words? How can the symbol grounding mechanism be extended to generate and ground abstract categories?
- Dialogue and interactive alignment for cooperative decision making University of Plymouth, UK (Angelo Cangelosi, Anna-Lisa Vollmer)
Many of the processes involved in a communicative/cooperation tasks (e.g. intention recognition, mutual alignment, action coordination) are increasingly found to function automatically and be grounded in lower levels. New experiments on communication between robots, and humans and robots, will investigate how interactive alignment can be achieved during social collaboration on a shared problem solving and decision making activities (e.g. on a joint object manipulation/assembly task). What mechanism can be designed to control dynamic alignment capabilities?
- Quantitative framework for sensory-motor strategies Zurich University, Switzerland (Rolf Pfeifer, Naveen Kuppuswamy)
Recent research has shown that dynamical structures emerge from physical interactions among sensorymotor flows between the agent and its environment. We hypothesise that the induced information structure can be incorporated in the sensory-motor representation of the robot’s body (loosely speaking, its “body schema”) and will serve as a basis for a forward model and predictive control (e.g. what sensory feedback to expect when grasping an object “on the fly”). This project will employ quantitative tools based on non-linear dynamics and information theory to investigate how to systematically exploit various properties of a robot’s body and achieve the formation of the body schema through the integration and causal analysis of crossmodal sensory-motor experiences.
- Exploitation of biomechanical constraints for learning of complex movements Zurich University, Switzerland (Rolf Pfeifer, Cristiano Alessandro)
Studies in biomechanics have shown that the amazingly robust motor skills in animals are the results of an interplay between clever biomechanical structures and adaptive sensory-motor control. This project will investigate how to exploit biomechanical constraints of the robot’s body in generating the appropriate sensory-motor strategies for complex movements. We will explore this problem in two ways, i.e. by simulating the constraints in software on existing platforms and by implementing a multi-DOF arm with variable compliance actuation. We will then utilize these biomechanical constraints to develop and utilize neuro-mechanical (periodic and discrete) motor primitives for generating and learning complex and adaptiverobotic movements.
- Motoric and contextual information in determining attention behaviours Italian Institute of Technology, Italy (Giorgio Metta, ER to be emplyoed)
- Motoric and contextual information in object recognition Italian Institute of Technology, Italy (Giorgio Metta, Nick Wilkinson)
This line of research will address the problem of analyzing, modelling and building systems that can recognize objects by including contextual and motoric information. This requires a new way of looking at the problem of learning to extract relevant information at each stage of processing (e.g. how to determine the visual features? What is the relationship between vision and touch in object recognition? How motor information structures the recognition process?). Of particular interest is the determination of the link between object-related information and the pragmatic use of this information for the control of manipulation.
- Affective ‘representation’ and anticipation Skovde University, Sweden (Tom Ziemke, Robert Lowe, Kiril Kiryazov)
Inspired by Damasio’s notion of “feelings” as “brain representations” of emotional bodily states and Grush’s emulation theory of representation, the following questions will be investigated, among others: How can an embodied agent make use of its perception/interoception of own body-internal/homeostatic states (in addition to perception of the external environment) in structuring its cognition and behaviour? In particular, how can embodied simulation/anticipation of both sensorimotor and affective consequences of actions interact in the preparation and planning of action (possibly at different levels of abstraction)?
- Affective modulation of embodied cognition Skovde University, Sweden (Tom Ziemke, Robert Lowe, Gauss Lee)
This research will expand on the ICEA project’s cognitive-affective architecture involving different levels of homeostatic regulation in embodied cognitive agents, based on neurocomputational models of the interaction of different mammalian brain structures. While the work in ICEA has been based mainly on models of rat behaviour, the new research will be adapted to the more complex iCub humanoid platform as well as more human-like cognition and behaviour. Research questions include: How can (higher levels of) cognition be grounded in and informed by affective mechanisms? In particular, how can decisionmaking/ action selection be based on so-called somatic-markers. How do the time scales of processes at different levels interact?
- Development of dialogical rules Bielefeld University, Germany (Britta Wrede, Vikram Narayan)
The verbal and cognitive development of infants is rooted in dialogue with other people (e.g. parents, peers). However, there is little research on how infants develop the capability for dialogue. We hypothesise that contingency is a fundamental mechanism that may help infants to develop their basic interactive capabilities such as turn-taking. These capabilities guide their attention to opening phases of a dialogue, and even regulate their emotions. This project proposes to use the mechanism of contingency to build a system that can learn dialogical rules through interaction by analyzing the effects of its own dialogue contributions.
- Emergence of top-down strategies from bottom-up social attentional processes Bielefeld University, Germany (Katharina Rollfing, Joanna Znajdek)
The development of top-down preferences for social cues helps infants to establish an enduring focus on, and give high priority to, the relevant information through its integration with saliency-based bottom-up processes. Here we aim at investigating how such top-down strategies can be derived from basic mechanisms and how they require and advance system memory and learning capabilities. Integrating such a capability on a robot would enable the study of the influence of the environment and the history of social interaction on the emerging top-down processes, and allow the robot to gradually develop more focused topdown attentive behaviours shaped by joint social and bottom-up cues.
- Neural Emotion and Action Integration Hamburg University, Germany (Stefan Wermter, Nicolas Navarro-Guerrero)
Mirror neurons become active not only when a primate performs an action, but also when it observes or is instructed to perform said action. So far these neurons have been seen in Broca’s and premotor cortical areas of the brain, but a recent study has found similar neurons in the insula of the brain, a new area with autonomic and gustatory functions. Neuroscientific evidence indicates that the mirror neurons are not only relevant for action understanding but also are linked to emotions. We propose to develop associative reinforcement neural networks for integrating emotional, tactile, visual, and acoustic input.
- Neural Cognitive Integration Hamburg University, Germany (Stefan Wermter, Junpei Zhong)
Mammals and humans perform well for tracking objects under difficult conditions. While robots can perform tracking using traditional methods, under difficult noisy conditions learning and multimodal integration can support visual tracking. Important research questions are: How can a brain-inspired low level system be modelled? What role does the superior colliculus play for map alignment? How are saccades and head
- Infants’ learning of manual control for the exploration and handling of objects Uppsala University, Sweden (Gustaf Gredebäck, Andrea Handl)
New research will look at what kind of object information infants use in the planning of actions and how they use it to improve the smoothness and controllability of the movements. Investigations will focus on the infant’s knowledge and use of the functional properties of objects in the planning of actions. The aim is to study how infants visually fixate different parts of objects before they reach for them and how they fixate objects during their manual exploration of them.
- Infants’ understanding of other people’s actions. Uppsala University, Sweden (Gustaf Gredebäck, Claudia Elsner)
In the same way as adults, infants fixate the goals of their actions ahead of time. When they reach for an object, they fixate the object before moving the hand there. Furthermore, infants fixate the goals of other people’s actions before they are fully implemented. We will use this paradigm to study how infants develop their understanding of other people’s actions.
- Bodyware technology for the development of cognition Telerobot Ocem Srl, Italy (Francesco Becchi, Giovann Stellin, Wiktor Sieklicki)
Since motoric and bodily information shapes the development of cognitive behaviours, it is important to investigate how to design new mechatronic systems dedicated to the study of cognition. Though this investigation has been started with the realisation of the iCub robot, several aspects still require improvement such as joint-level force control, weight distribution optimisation, better tracking and control of the oculomotor system, active and passive compliance, and full sensorisation. This line of research will range from mechatronic design to control strategies involving new sensors, with the aim of exploiting biologically inspired principles (e.g. muscle like control) in the design of cognitive robots. For these reasons new materials and manufacturing technologies will also be explored.