Current Research Projects

 

 

Joint Action Science and Technology (JAST)

financed by the European Commission, FP6 IST-2-003747-IP.

 

The JAST consortium brings together research groups from cognitive science and robotics. The main objective is to build jointly acting autonomous systems that work intelligently on mutual tasks. To realise this objective, JAST exploits a prototypical research paradigm in which two autonomous agents (human-human, robot-human, robot-robot) perform a single construction task. The different experimental groups involved cover in their research all aspects of joint action including the perceptual, reasoning and motor level.  The results of the neuro-cognitive experiments are used by our group  to develop integrative cognitive models for joint action. They are based on the theoretical framework of dynamical systems and dynamic fields.  Dr. Estela Bicho’s group at the Department of Industrial Electronics, University of Minho, is responsible for the design and control of two mobile robotics platforms, each equipped with a 7-degrees-of-freedom arm, a three-fingered hand and a vision system. The dynamic control architectures are tested in various experimental scenarios which are part of the joint construction task. They emphasize different challenges for a robot team with minimal explicit communication such as for instance the timing and synchronisation of forces in joint transportation, the reasoning about the partner’s current action goal and the joint planning to achieve a desired end state (for more details see http://www.euprojects-jast.net/).

For a recent video of the Minho robot in joint action with a user see http://cordis.europa.eu/ictresults/

 

 

Consortium leader:  Dr. Harold Bekkering,  Donders Institute for Brain, Cognition and Behaviour, Centre for Cognition Radboud University Nijmegen, The Netherlands

___________________________________________________________________

 

 

Analysis of pattern formation in dynamic fields with bi-stable elements

financed by the Portuguese Science Foundation (FCT)

 

The main objective is to analytically and numerically investigate a class of integro-differential equations which consists of bi-stable elements and linear spatial interactions. In a certain parameter regime, these equations show the same dynamics of pattern formation like the “classical” dynamic fields of Amari type. In general, however, the dynamics differs systematically from conceptually related systems (e.g., Amari, RD-systems), which makes the system attractive for certain applications. Of particular interest are the dependence of the shape of localized patterns on the initial state, the robustness of stationary pulse solutions against perturbations of the interaction symmetry and the co-existence of multiple pulse solutions. 

 

 

 

 

Learning to read the motor intention of others: towards socially intelligent robots,

financed by FCT (POCI/V.5/A0119/2005)

 

The main objective of the project is to further develop our dynamic field model for action understanding which we have originally proposed in the contetxt of  goal-directed imitation. To this end, we analyze and interpret  neural data recorded in dedicated monkey experiments by our partners at the University of Parma, Italy. The STS-PF/PFG-F5 mirror circuit is considered to play a fundamental role in the capacity to understand and imitate actions of others.  The defining functional characteristic of mirror neurons is that they become active not only when the monkey executes a particular motor act (like grasping, holding or placing  an object) but also when it observes another individual (monkey or human) performing a similar motor act. Recent findings by Fogassi and colleagues (Science, 308: 662-667, 2005) show that  neurons in the monkey's   inferior parietal lobule,  while coding a specific motor act, show markedly different activation patterns depending upon the final goal of the action sequence in which the act is embedded. Our working hypothesis for the modeling the intention reading capacity is that neurons in parietal cortex are organized in chains of motor acts dedicated to achieve certain action goals. Important cues for triggering the motor chains are visual inputs representing object properties or observed motor acts (e.g., the first primitive of a chain).

The emphasis of the project is on the developmental aspect: we apply learning dynamics to establish during “practice” connections between dynamics field representations which eventually build the goal-directed chains. 

 

Project partners:

Dr. Leonardo Fogassi and Dr. Pier F. Ferrari, University of Parma, Italy

Dr. Albert Mukovskiy, Institute of Higher Nervous Activity and Neurophysiology, Russ. Acad. Sci., Russia

 

________________________________________________________________

 

Modeling the dynamics of activity patterns in primary visual cortex

financed by CRUP: Acções Integradas Luso-Alemãs

 

The primary visual cortex is the most intensively investigated cortical tissue in neurobiology. However, its contribution to sensation is still under debate. Dirk  Jancke and colleagues (Nature 438: 423-426, 2004)  have recently demonstrated the impact of spatio-temporal patterns of subthreshold synaptic potentials on early cortical processing and hence its potential contribution on the shaping of perception.

Our goal in this project is to identify the neuronal interactions underlying these spatio-temporal patterns through dynamical models adapted to data from optical imaging and single unit recordings. The data will be interpreted quantitatively using a dynamic neural field approach that provides theoretically meaningful parameters (average coupling strengths, time constants etc.)  directly linked to the functional cortical architecture.  By systematically adapting these models to the experimental data using evolutionary algorithms, we may identify these model parameters. Mathematical analysis of the resulting model may reveal its principle information processing capabilities.

A second goal of the project is to link the neuronal and the perceptual level. Following our previous work we use the dynamic field models to hypothesize about neural processing mechanisms underlying systematic localization errors which occur with objects in motion (e.g., line motion, Fröhlich effect, flash-lag effect, representational momentum).

 

Project partners:

Dr. Dirk Jancke, Neurobiology, Ruhr-University Bochum, Germany

Dr. Christian Igel, Neuroinformatics, Ruhr-University Bochum, Germany

 

 

 

Synthesis of Cooperative Behavior in Multi-robot Systems:

a Nonlinear Attractor Dynamics Approach (COOPDYN)

financed by the Portuguese Science Foundation (FCT)

 

The main goal is to design and test control architectures for cooperative behavior among mobile robots using the dynamic approach to behavior generation. The robot platforms are equipped with low-level sensors, the explicit communication between the robots is minimal (e.g., through forces) and there is now prior knowledge about the world.  The experimental scenarios include for instance the motion in formation and the joint transportation of objects. The experiments with the real robots give insights about the strength and also the limitations of sensor-near approach to cooperative behavior.

 

Project leader: Dr. Estela Bicho, DEI, University of Minho, Portugal