Übersetzung in Arbeit!

Perceptual Learning, Proprioception and Body Image

In the well-konwn Rubber Hand Illusion (RHI), Botvinick and Cohen (1998) have shown that through several minutes of synchronous tactile stimulation of the invisible own hand and a visible rubber hand, the illusion of feeling stimulation in the fake hand can be evoked. This strking illusion relies only on patterns of visuotactile coherence and is marked by two related but different percepts, i.e., the illusion of ownership of the rubber hand and a proprioceptive drift of perceived hand position towards the rubber hand. Follow-up work (e.g., Tsakiris and Haggard, 2005; Armel and Ramachandran, 2003) has investigated this effect and identified a set of counter-intuitive results on conditions for stability and occurrence of the effect, e.g., that it is very sensitive to delays in otherwise coherent stimulation, but very robust to perturbations of visual appearance of the hand, such as colour, shape and size. We are using a fully automatised set-up with two PHANToM Force Feedback devices to investigate different dimensions of this interesting phenomenon, in particular the illusion of proprioceptive drift in a more detailed manner. We look at levels of noise, the role of action in the illusion and statistical properties of the effect across and within subjects, trying to identify patterns of statistical optimality or other formal rules governing the occurrence and magnitude. This illusion, which is not just visuotactile but is also influenced by top-down cues and motion, reveals fundamental mechanisms of self-appropriation, perceived unity and the construction of identity and body image and, therefore, has implications for work in many areas, such as human computer interfaces, prosthetics and neural disorders of body scheme and body image. Understanding the rubber hand illusion also is an important contribution towards understanding the mechanisms of perceptual learning and and multimodal integration in more general terms, which is an area that we still know little about.

Adaptation to Sensory Delays and Perceived Simultaneity

Time and temporality are amongst the most fundamental dimensions of our cognitive world and the experiments, models and theories about time cognition are countless. I have been working on time cognition in general and on the problem of perceived simultaneity in particular. I have been particularly intrigued by recent results on how adaptation to sensory delays can induce a shift of perceived simultaneity, and how this adaptation can even lead to a reversal of perceived order of events if the delay is subsequently removed (e.g., Cunningham et al., 2001). These results are in direct opposition to earlier results on adaptaiton to delay that reported little to no plasticity in perceived simultaenity and led scientists to conclude that in principle impossible (Smith and Smith,1962). My work on this problem was interdisciplinary, including an experimental study of adaptation to temporal delay (chapters 9 and 11 of my D.Phil. Dissertation, see also Rohde and Gapenne, 2006) which I conducted during a research placement in the CRED, UT Compiegne, an Evolutionary Robotics simulation of the same problem ( Rohde and Di Paolo, 2008; chapter 10 of my D.Phil. dissertation) and conceptual work on the phenomenology and sensorimotor basis of time perception (chapter 8 of my D.Phil. Dissertation). The experimental results in themselves were not conclusive, but the overall project resulted in both, a number of concrete hypotheses for further experimentation in order to establish the exact conditions under which adaptation to sensory delays can occur and also a far-reaching perspective about the sensorimotor basis of time perception and temporal cognition in general that suggests a larger scale research program (chapter 12 of my D.Phil. dissertation).

Enaction, Constructivism and Methodological Issues

Cognitive Science has in recent years experienced a shift away from the original more logics based computational paradigm towards more embodied and dynamical approaches that take into consideration emergent effects from sensorimotor interaction with the environment and the role of morphology, motion and ecological viability for how our cognitive and perceptual world is structured. I am a strong advocate of this kind of approach, in particular the enactive (Varela, Thompson and Rosch, 1991; Stewart, Gapenne, & Di Paolo, forthcoming) approach and have long since taken an interest in science-theoretic and methodological issues implicated by such a shift in focus. For me, this form of non-reductive naturalistic world-view, away from a computationalist-representationalist theory of mind implies an epistemological constructivism. My D.Phil. dissertation, which investigates the use of Evolutionary Robotics simulation models in the study of human mind and cognition deals in large parts with promoting and developing this approach and its methodological assets (in particular, chapters 2, 3 and 13). I propose a minimalist interdisciplinary sensorimotor approach that studies perceptual phenomena using the methods of experimental psychophysics and situated simulation modelling, relating empirical measurement to perceptual experience and abstract formal principles of sensorimotor organisation. Large parts of my work on the enactive approach as a general explanatory framework have been published in Di Paolo, Rohde and De Jaegher (forthcoming). I also argue for scientific explanation in Cognitive Science that strives towards explanation of generative mechanisms without reducing cognitive phenomena to the mechanisms explained. In Rohde and Stewart 2008, we develop in more detail how such an approach achieves to escape the fallacies of either eliminativism or functional reductionism by making explicit that the scientist herself is a subject and how this reflects in her activity of scientific knowledge construction.

Perceptual Crossing in Simulated Environments

Enactive approaches to social cognition advocate a global and interaction centred approach to problems of social cognition, rather than one that centres around the individual and its cognitive capacities in isolation. Lenay et al. (e.g., Auvray, Lenay and Stewart, 2008) have developed an experimental paradigm to study agency detection and perceptual crossing in a minimalist simulated environment that exemplifies how the dynamics of embodied interaction processes can give rise to complex and coordinated behaviours that go far beyond what simple agents could achieve due to their individual cognitive faculties were they in isolation. Blindfolded subjects are asked to identify the other participant and distinguish it from distractor items (static object, shadow image of the other) through tactile feedback. Their results show how human subjects reliably achieve rhythmic interaction with one another as an emergent property of the multi-agent-environment interaction, even though such agency detection goes beyond individual cognitive discriminative capacity (i.e. perceptual judgment shows incapacity to cognitively distinguish the shadow image from the other). The task and the simple simulated environment used in the study are so minimalist that simple simulated agents can be evolved to perform an identical task. I have thus modelled the original experiment (Di Paolo, Rohde and Iizuka, 2008, chapter 6 of my D.Phil. dissertation) as well as its extension to a two-dimensional environment (Rohde and Di Paolo, 2008; chapter 7 of my D.Phil. dissertation). The results of this simulation model do not only confirm the analysis provided, but also point out further issues that have been previously overlooked, such as the sensorimotor similarity of interaction with the other or a static object. Moreover, the simulation generates a number of quantitative predictions about certain aspects of the sensorimotor dynamics, i.e., that integrated stimulation time is a formal correlate of the behavioural distinction between the mobile and static attractors in the simulated environment, or, in the two-dimensional variant, that arm morphology plays a role in the quantitative expression, but not the general logic of the strategies observed in human subjects. This cunningly simple set-up has proven to be very rich and modifications, extensions and replications are now being investigated by several groups working within sensorimotor approaches, both experimentally (Lenay et al., partially unpublished work; Di Paolo, De Jaegher and Wood, unpublished work) and in simulation (Froese and Di Paolo, 2008; Martius et al., 2008) and provides a good example for how simple simulation models can complement experimental approaches to perception.

Evolutionary Robotics and the Emergence of Function and Meaning

The most exciting project for an enactivist/constructivist is to study how meaning is constructed by humans through their interaction with others, the environment, their culture, their body, their developmental history etc. However, it is sometimes equally interesting to take a look at simple agents, bacteria, animals, and also simulated or real robots, to get an idea of how such processes of dynamic and embodied self-organisation can work in principle, what is the same across systems, and what is different. Evolutionary Robotics is a valuable tool for this kind of "mental gymnastics" (cf. Harvey et al. 2005), to provide proofs of concept and counter our intuitions. I have been using Evolutionary Robotics models in this spirit to investigate the implicit prior assumptions underlying so called "hybrid approaches", i.e., those models that aim to combine more embodied and more traditional ways of modelling (cf. Rohde and Di Paolo, 2006, also chapter 5 of my D.Phil. dissertation). I use the example of self-supervised learning architectures in this study, but the point made extends to hybrid architectures more generally, I think. This research connects to a more ambitious exploration into the origins of value and meaning in biological organisms as we describe in more detail in Di Paolo et al. (forthcoming), which takes inspiration from the philosophy of biology and looks at self-maintaining processes of identity generation at different levels (cellular, multi-cellular, neural, social, ...) and at different stages of phylogenetic evolution. The idea is that, in order to overcome the functional limitations associated with localised and symbolic meaning generation, modelling will have to try and capture some of these dynamics of self-preservation.

Neural Dynamics and Embodied Behavior

The dynamical properties of our brains clearly impact on and constrain the kinds of behaviours we can exhibit. I have always been fascinated by approaches that try to link the dynamical properties that characterise real or artificial neural networks to the realisation of a certain behaviour in closed loop interaction with the environment. Recently, I have been working with J.-J. Aucouturier and T. Ikegami at the University of Tokyo on an extension of their work on chaotic itinerancy to bring about interesting and pleasant dancing behaviour in a commercial entertainment dancing robot MIURO. In their earlier study (Aucouturier, J.J., Ogai, Y. and Ikegami, T., 2007), the group used FitzHugh Nagumo neural networks which are characterised by being prone to complex dynamics, in order to generate a controller that responds in a chaotic manner to a rhythmic input signal, exhibiting alternatively phases of rhythmical coupling (in varying patterns) and phases of decoupling from the rhythm fed in, which is a pattern similar to the one we find in dancing humans. We use evolutionary techniques to assess, quantify and control this dynamics. The ultimate aim of this study is to investigate if these kinds of dynamics, which are currently only implemented in the open loop, can also lead to interesting and life-like patterns in the closed loop. From 2004-2006, I had been coordinating the reading group DyStURB - Dynamical Structures to Understand Real Brains in order to work on these kinds of issues. It started out as a reading group strictly related to neural dynamics, but turned more and more into a general forum to discuss ideas related to enactive and dynamical approaches. In the MSc term paper Evolving Spiking Neural Networks for Robot Control. I replicated and extended work on linking spike time dependent plasticity in evolved agents to functional properties of the behaviour evolved. Finally, in my BSc dissertation on Dynamical Properties of Self-Regulating Neurons I looked at the bifurcation properties of a certain type of neural network with homeostatic synaptic plasticity in the open loop. These dynamics that were later linked to behavioural stability in closed loop evolved behaviour. This research was generated during my placement in the Intelligent Dynamical Systems (INDY) group at the Fraunhofer AIS in St. Augustin under the supervision of Prof. Pasemann. I presented a poster on this work at the IK2003, and these results have been included in a paper that has been submitted to neural networks (see abstract).

Motor Synergies as Principles in Motor Control

A project during my D.Phil. (following up on my MSc dissertation) was the exploration of the concept of motor synergies (Bernstein, 1967) as a principle in motor control (Rohde and Di Paolo, 2005). Inspired by an experimental physiological study (Gottlieb et. al., 1997), I evolved robot controllers for a directional pointing task. I found that linear synergies, even though they do not seem to be necessary in order to generate successful behaviour in a reaching task, greatly enhanced the evolvability of the behaviour. These findings are in line with another experimental study reporting that linear synergies constrain the development of reaching movements in human infants, rather than being acquired as the outcome of motor development ( Zaal et al., 1999). The simulation was cited as supportive evidence in a follow-up study by the original group (Shemmell et al., 2007) and we were encouraged to perform more conceptual testing in silicio.