Dynamic Neural Fields

I am indebted to many people for their contribution to my knowledge and for their help in many ways during these past years. I wish to express my sincere gratitude to my scientific adviser Dr. Gregor Schöner, from the CRNC-CNRS in Marseille, who initiated me into science and who has accompanied me during the past years. I had the pleasure to experience and profit from his brilliance, his friendship and his unconventional ideas. Furthermore, traveling and discussing with him around the world was a deep pleasure. I want to thank Prof. Francisco Vaz, from the Department of Electronics and Telecommunications at the University of Aveiro, for his humanity and for making my scientific life possible at first place. Thanks are also due to Prof. Carlos Couto, from the Department of Industrial Electronics at the University of Minho, for his interest and continuing institutional support. I wish to express my respect to Prof. Keith Doty and Scott Jantz, from the Machine Intelligence Laboratory at the University of Florida in Gainsville, for their knowhow in building autonomous robots. I am indebt for their contribution to my knowledge in behavior-based robotics. Pierre Mallet, who worked with me in the "Equipe de Dynamique" at the CRNC-CNRS in Marseille, deserves my deepest thanks for constructing Robodyn and for his continuing support and confidence. I also thank him for his careful reading of the manuscripts. I am grateful to Prof. von Seelen for his generous hospitality at the "Institut für Neuroinformatik" in Bochum. The stays in his institute and the interactions with my "dynamical peers" were fundamental for the final phase of my thesis. I am indeed indebted to Wolfram Erlhagen, Rainer Menzner and Axel Steinhage for careful proof-reading, for their interest and their helpful comments on this thesis. They were always there to help in any situation. Thanks also due to Martin Giese. My thanks also go to Michael Neef for his quick and excellent assistance in technical issues. I am grateful to the members of the "Dynamikgruppe" for making my participation possible in the Friday meetings. At the CRNC-CNRS in Marseille I would like to express my gratitude to Dr.

I am indebted to many people for their contribution to my knowledge and for their help in many ways during these past years. I wish to express my sincere gratitude to my scientific adviser Dr. Gregor Schöner, from the CRNC-CNRS in Marseille, who initiated me into science and who has accompanied me during the past years. I had the pleasure to experience and profit from his brilliance, his friendship and his unconventional ideas. Furthermore, traveling and discussing with him around the world was a deep pleasure. I want to thank Prof. Francisco Vaz, from the Department of Electronics and Telecommunications at the University of Aveiro, for his humanity and for making my scientific life possible at first place. Thanks are also due to Prof. Carlos Couto, from the Department of Industrial Electronics at the University of Minho, for his interest and continuing institutional support. I wish to express my respect to Prof. Keith Doty and Scott Jantz, from the Machine Intelligence Laboratory at the University of Florida in Gainsville, for their knowhow in building autonomous robots. I am indebt for their contribution to my knowledge in behavior-based robotics. Pierre Mallet, who worked with me in the "Equipe de Dynamique" at the CRNC-CNRS in Marseille, deserves my deepest thanks for constructing Robodyn and for his continuing support and confidence. I also thank him for his careful reading of the manuscripts. I am grateful to Prof. von Seelen for his generous hospitality at the "Institut für Neuroinformatik" in Bochum. The stays in his institute and the interactions with my "dynamical peers" were fundamental for the final phase of my thesis. I am indeed indebted to Wolfram Erlhagen, Rainer Menzner and Axel Steinhage for careful proof-reading, for their interest and their helpful comments on this thesis. They were always there to help in any situation. Thanks also due to Martin Giese. My thanks also go to Michael Neef for his quick and excellent assistance in technical issues. I am grateful to the members of the "Dynamikgruppe" for making my participation possible in the Friday meetings. At the CRNC-CNRS in Marseille I would like to express my gratitude to Dr.

We present a control architecture for non-verbal HRI that allows an assistant robot to have a pro-active and anticipatory behavior. The architecture implements the coordination of actions and goals among the human, that needs help, and the robot as a dynamic process that integrates contextual cues, shared task knowledge and predicted outcome of the human motor behavior. The robot control architecture is formalized by a coupled system of dynamic neural fields representing a distributed network of local but connected neural populations with specific functionalities. Different subpopulations encode task relevant information about action means, action goals and context in form of self-sustained activation patterns. These patterns are triggered by input from connected populations and evolve continuously in time under the influence of recurrent interactions. The dynamic control architecture is validated in an assistive task in which an anthropomorphic robot acts as a personal assistant of a person with motor impairments. We show that the context dependent mapping from action observation onto appropriate complementary actions allows the robot to cope with dynamically changing situations. This includes adaptation to different users and mutual compensation of physical limitations. The present research was conducted in the context of the fp6-IST2 EU-project JAST (proj.nr. 003747) and partly financed by the FCT grant POCI/V.5/A0119/2005.

Programming by demonstration allows non-experts in robot programming to train the robots in an intuitive manner. However, this learning paradigm requires multiple demonstrations of the same task, which can be time-consuming and annoying for the human tutor. To overcome this limitation, we propose a fast learning system-based on neural dynamics-that permits collaborative robots to memorize sequential information from single task demonstrations by a human-tutor. Important, the learning system allows not only to memorize long sequences of sub-goals in a task but also the time interval between them. We implement this learning system in Sawyer (a collaborative robot from Rethink Robotics) and test it in a construction task, where the robot observes several human-tutors with different preferences on the sequential order to perform the task and different behavioral time scales. After learning, memory recall (of what and when to do a sub-task) allows the robot to instruct inexperienced human workers, in a particular human-centered task scenario.

We report on our approach towards creating socially intelligent robots, which is heavily inspired by recent experimental findings about the neurocognitive mechanisms underlying action and emotion understanding in humans. Our approach uses neuro-dynamics as a theoretical language to model cognition, emotional states, decision making and action. The control architecture is formalized by a coupled system of dynamic neural fields representing a distributed network of local but connected neural populations. Different pools of neurons encode relevant information in the form of self-sustained activation patterns, which are triggered by input from connected populations and evolve continuously in time. The architecture implements a dynamic and flexible context-dependent mapping from observed hand and facial actions of the human onto adequate complementary behaviors of the robot that take into account the inferred goal and inferred emotional state of the co-actor. The dynamic control architec...

In this paper we present a model for action preparation and decision making in cooperative tasks that is inspired by recent experimental findings about the neuro-cognitive mechanisms supporting joint action in humans. It implements the coordination of actions and goals among the partners as a dynamic process that integrates contextual cues, shared task knowledge and predicted outcome of others' motor behavior. The control architecture is formalized by a system of coupled dynamic neural fields representing a distributed network of local but connected neural populations. Different pools of neurons encode task-relevant information about action means, task goals and context in the form of self-sustained activation patterns. These patterns are triggered by input from connected populations and evolve continuously in time under the influence of recurrent interactions. The dynamic model of joint action is evaluated in a task in which a robot and a human jointly construct a toy object. We show that the highly context sensitive mapping from action observation onto appropriate complementary actions allows coping with dynamically changing joint action situations.

In this paper we present results of our ongoing research on non-verbal human-robot interaction that is heavily inspired by recent experimental findings about the neuro-cognitive mechanisms supporting joint action in humans. The robot control architecture implements the joint coordination of actions and goals as a dynamic process that integrates contextual cues, shared task knowledge and the predicted outcome of the user's motor behavior. The architecture is formalized by a coupled system of dynamic neural fields representing a distributed network of local but connected neural populations with specific functionalities. We validate the approach in a task in which a robot and a human user jointly construct a toy 'vehicle'. We show that the context-dependent mapping from action observation onto appropriate complementary actions allows the robot to cope with dynamically changing joint action situations. This includes a basic form of error monitoring and compensation. 1 The present research was conducted in the context of the fp6-IST2 EUproject JAST (proj.nr. 003747) and partly financed by the FCT grants POCI/V.5/A0119/2005 and CONC-REEQ/17/2001.

We report on our approach towards creating socially intelligent robots, which is heavily inspired by recent experimental findings about the neurocognitive mechanisms underlying action and emotion understanding in humans. Our approach uses neuro-dynamics as a theoretical language to model cognition, emotional states, decision making and action. The control architecture is formalized by a coupled system of dynamic neural fields representing a distributed network of local but connected neural populations. Different pools of neurons encode relevant information in the form of self-sustained activation patterns, which are triggered by input from connected populations and evolve continuously in time. The architecture implements a dynamic and flexible context-dependent mapping from observed hand and facial actions of the human onto adequate complementary behaviors of the robot that take into account the inferred goal and inferred emotional state of the co-actor. The dynamic control architec...

LORIA-Campus Scientifique-BP 239-54506 Vandoeuvre-les-Nancy Cedex (Received 00 Month 200x; in final form 00 Month 200x) In [26], we've introduced a dynamic model of visual attention based on the Continuum Neural Field Theory [29] that explained attention as being an emergent property of a dynamic neural field. The fundamental property of the model is its facility to select a single stimulus out of several perfectly identical input stimuli by applying asynchronous computation. In the absence of external noise and with a zero initial state, the theoretical mathematical solution of the field equation predicts the final equilibrium state to equally represent all of the input stimuli. This finding is valid for synchronous numerical computation of the system dynamics where elements of the spatial field are computed all together at each time point. However, asynchronous computation, where elements of the spatial field are iterated in time one after the other yields different results leading the field to move towards a single stable input pattern. This behavior is in fact quite similar to the effect of noise on dynamic fields. The present work aims at studying this phenomenom in some details and characterizes the relation between noise, synchronous evaluation (the "regular" mathematical integration) and asynchronous evaluation in the case of a simple dual particle system. More generally, we aim at explaining the behavior of a general differential equation system when it is considered as a set of particles that may or may not iterated by synchronous computations.

The visual exploration of a scene involves the interplay of several competing processes (for example to select the next saccade or to keep fixation) and the integration of bottom-up (e.g. contrast) and top-down information (the target of a visual search task). Identifying the neural mechanisms involved in these processes and in the integration of these information remains a challenging question. Visual attention refers to all these processes, both when the eyes remain fixed (covert attention) and when they are moving (overt attention). Popular computational models of visual attention consider that the visual information remains fixed when attention is deployed while the primates are executing around three saccadic eye movements per second, changing abruptly this information. We present in this paper a model relying on neural fields, a paradigm for distributed, asynchronous and numerical computations and show that covert and overt attention can emerge from such a substratum. We identify and propose a possible interaction of four elementary mechanisms for selecting the next locus of attention, memorizing the previously attended locations, anticipating

This article was published in an Elsevier journal. The attached copy is furnished to the author for non-commercial research and education use, including for instruction at the author's institution, sharing with colleagues and providing to institution administration. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier's archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright

Programming by demonstration allows non-experts in robot programming to train the robots in an intuitive manner. However, this learning paradigm requires multiple demonstrations of the same task, which can be time-consuming and annoying for the human tutor. To overcome this limitation, we propose a fast learning system-based on neural dynamics-that permits collaborative robots to memorize sequential information from single task demonstrations by a human-tutor. Important, the learning system allows not only to memorize long sequences of sub-goals in a task but also the time interval between them. We implement this learning system in Sawyer (a collaborative robot from Rethink Robotics) and test it in a construction task, where the robot observes several human-tutors with different preferences on the sequential order to perform the task and different behavioral time scales. After learning, memory recall (of what and when to do a sub-task) allows the robot to instruct inexperienced human workers, in a particular human-centered task scenario.

Many of the tasks we perform during our everyday lives are achieved through sequential execution of a set of goal-directed actions. Quite often these actions are organized hierarchically, corresponding to a nested set of goals and subgoals. Several computational models address the hierarchical execution of goaldirected actions by humans. However, the neural learning mechanisms supporting the temporal clustering of goal-directed actions in a hierarchical structure remain to a large extent unexplained. In this paper we investigate in simulations, of a dynamic neural field (DNF) model, biologically-based learning and adaptation mechanisms that can provide insight into the development of hierarchically organized internal representations of naturalistic tasks. In line with recent experimental evidence from observational learning studies, the DNF model implements the idea that prediction errors play a crucial role for grouping fine-grained events into larger units. Our ultimate goal is to use the model to endow the humanoid robot ARoS with the capability to learn hierarchies in sequential tasks, and to use that knowledge to enable efficient collaborative joint tasks with human partners. For testing the ability of the system to deal with the real-time constraints of a learning-by-demonstration paradigm we use the same assembly task from our previous work on human-robot collaboration. The model provides some insights on how hierarchically structured task representations can be learned and on how prediction errors made by the robot and signaled by the demonstrator can be used to control such process.

We present a control architecture for non-verbal HRI that allows an assistant robot to have a pro-active and anticipatory behavior. The architecture implements the coordination of actions and goals among the human, that needs help, and the robot as a dynamic process that integrates contextual cues, shared task knowledge and predicted outcome of the human motor behavior. The robot control architecture is formalized by a coupled system of dynamic neural fields representing a distributed network of local but connected neural populations with specific functionalities. Different subpopulations encode task relevant information about action means, action goals and context in form of self-sustained activation patterns. These patterns are triggered by input from connected populations and evolve continuously in time under the influence of recurrent interactions. The dynamic control architecture is validated in an assistive task in which an anthropomorphic robot acts as a personal assistant of a person with motor impairments. We show that the context dependent mapping from action observation onto appropriate complementary actions allows the robot to cope with dynamically changing situations. This includes adaptation to different users and mutual compensation of physical limitations. The present research was conducted in the context of the fp6-IST2 EU-project JAST (proj.nr. 003747) and partly financed by the FCT grant POCI/V.5/A0119/2005.

In this paper we present a model for action preparation and decision making in cooperative tasks that is inspired by recent experimental findings about the neuro-cognitive mechanisms supporting joint action in humans. It implements the coordination of actions and goals among the partners as a dynamic process that integrates contextual cues, shared task knowledge and predicted outcome of others' motor behavior. The control architecture is formalized by a system of coupled dynamic neural fields representing a distributed network of local but connected neural populations. Different pools of neurons encode task-relevant information about action means, task goals and context in the form of self-sustained activation patterns. These patterns are triggered by input from connected populations and evolve continuously in time under the influence of recurrent interactions. The dynamic model of joint action is evaluated in a task in which a robot and a human jointly construct a toy object. We show that the highly context sensitive mapping from action observation onto appropriate complementary actions allows coping with dynamically changing joint action situations.

Programming by demonstration allows non-experts in robot programming to train the robots in an intuitive manner. However, this learning paradigm requires multiple demonstrations of the same task, which can be time-consuming and annoying for the human tutor. To overcome this limitation, we propose a fast learning system-based on neural dynamics-that permits collaborative robots to memorize sequential information from single task demonstrations by a human-tutor. Important, the learning system allows not only to memorize long sequences of sub-goals in a task but also the time interval between them. We implement this learning system in Sawyer (a collaborative robot from Rethink Robotics) and test it in a construction task, where the robot observes several human-tutors with different preferences on the sequential order to perform the task and different behavioral time scales. After learning, memory recall (of what and when to do a sub-task) allows the robot to instruct inexperienced human workers, in a particular human-centered task scenario.

Inhibition of return is characterized by delayed responses to previously attended locations when the cue-target onset asynchrony (CTOA) is long enough. However, when cues are predictive of a target’s location, faster reaction times to cued as compared to uncued targets are normally observed. In this series of experiments investigating saccadic reaction times, we manipulated cue predictability to 25% (counter-predictive), 50% (non-predictive), and 75% (predictive) to investigate the interaction between predictive endogenous facilitatory (FCEs) and inhibitory cueing effects (ICEs). Overall, larger ICEs were seen in the counter-predictive condition than in the non-predictive condition, and no ICE was found in the predictive condition. Based on the hypothesized additivity of FCEs and ICEs, we reasoned that the null ICEs observed in the predictive condition are the result of two opposing mechanisms balancing each other out, and the large ICEs observed with counter-predictive cueing can be attributed to the combination of endogenous facilitation at uncued locations with inhibition at cued locations. Our findings suggest that the endogenous activity contributed by cue predictability can reduce the overall inhibition observed when the mechanisms occur at the same location, or enhance behavioral inhibition when the mechanisms occur at opposite locations.

Inhibition of return (IOR) refers to an increase in reaction times to targets that appeared at a previously cued location relative to an uncued location, often investigated using a spatial cueing paradigm. Despite numerous studies that have examined many aspects of IOR, the neurophysiological mechanisms underlying IOR are still in dispute. The objective of the current research is to investigate the plausible mechanisms by manipulating the cue and target types between central and peripheral stimuli in a traditional cue-target paradigm with saccadic responses to targets. In peripheral cueing conditions, we observed inhibitory cueing effects across all cue-target onset asynchronies (CTOAs) with peripheral targets, but IOR was smaller and arose later with central targets. No inhibition was observed in central cueing conditions at any CTOAs. Empirical data were simulated using a two-dimensional dynamic neural field model. Our results and simulations support previous work demonstrating that, at short CTOAs, behavioral inhibition is only observed with repeated stimulation – an effect of sensory adaptation. With longer CTOAs, IOR is observed regardless of target type, when peripheral cueing is used. Our findings suggest that behaviorally exhibited inhibitory cueing effects can be attributed to multiple mechanisms, including both attenuation of visual stimulation and local inhibition in the superior colliculus.