PHD COURSE NEUROMORPHIC TACTILE SENSING
MARCH 25, ROOM 1
h. 10-11
Johan Wessberg, Institute of Neuroscience and Physiology, University of Gothenburg, Gothenburg, Sweden
Title: to be confirmed
Abstract: to be confirmed
h. 11-12
Henrik Jörntell, Neural Basis of Sensorimotor Control, Lund University, Lund, Sweden
Title: Brain processing of skin sensory information - From receptive field representations towards object representations
Abstract: Tactile sensory information is gained as our skin interacts with objects of the external world. The skin is endowed with an incredibly rich set of sensors, which transduce mechanical strains in the skin into patterns of neural spikes in the nerve fibers that convey the primary sensory information to the central nervous system. This presentation will be about how the primary sensory information is received and processed at the various processing stages within the hierarchically organized brain systems for touch-related information.
A fundamental component in all analysis of brain function is the fact that each neuron is a bioelectrical entity, which generates a potential difference across its cell membrane. This membrane potential can be varied in different ways to generate very fast electrical signals, by which the neurons communicate with each other. Although there are variations in the basic bioelectrical properties between different types of neurons, the similarities between most of the neurons of the brain is, in this respect, striking. It follows that the primary contribution to brain function lies in how the neurons are connected with each other (i.e. the network structure of the brain). Hence, to learn how the brain works it is crucial to learn how the neurons are connected with each other.
The first stage of processing is the cuneate nucleus of the brain stem. The neurons of the cuneate nucleus receive information from the primary sensory nerve fibers directly. This nucleus contains about 150,000 neurons, whereas the touch related information from the hand is carried by approximately 30,000 nerve fibers. Each nerve fiber fans out to contact about 1,500 cuneate neurons, resulting in that each cuneate neuron receives input from about 300 primary sensory fibers. Until recently, it was not known how the information from the sensors was integrated by the cuneate neurons, but a recent study in which this first stage of information processing was analyzed showed that each neuron has selected very few of the available primary sensory information, and to strengthen these synaptic inputs whereas synaptic inputs from other sources are weakened. This results in that the primary sensory information from a large number of nerve fibers are recombined in the cuneate neurons, and that each cuneate neuron can carry unique information. The cuneate nucleus also contains inhibitory interneurons, which depress the activity of the other neurons, the output neurons, of the cuneate nucleus. By way of the specific connectivity between the interneurons and the output neurons, this helps sharpening the contrast in the information conveyed. Notably, at this level, the brain still operates in terms of receptive fields, i.e. each neuron is only interested in information from the same skin area and there appears to be little integration of information across skin areas.
Information output by the cuneate neurons is provided to the thalamus, which subsequently conveys the information to the primary somatosensory cortex of our brains. The neurons of the thalamus and the somatosensory cortex have been investigated extensively, although not in as detailed fashion as in the cuneate nucleus. Briefly, the thalamus is a stage where the neocortex can select which information it is interested in, since the cortex projects directly to the thalamic neurons. Cortical information is integrated with the sensory information from the cuneate nucleus, before the information is conveyed to the neurons of the cortex. In the primary sensory cortex, the segregation of information from different receptive fields seem to be retained, as are different sensory ‘qualities’ (i.e. submodalitites). The cortex is a layered structure where each neuron layer is believed to make a specific contribution to the brain processing. This knowledge is not complete, however, and rarely extends beyond the first two input layers (which receive the information from the thalamus). Different cortical regions communicate with each other via the third layer, and information received from other cortical areas are received and integrated primarily by the neurons of the third and fifth layers. For our current understanding of the higher level representations of the sensory information, the analysis primarily stems from neuron s of the fifth layer. These representations reside in the second somatosensory cortex and the adjacent parietal association cortex, which correspond to a large portion of our brains. Some initial levels of analysis have suggested that the neurons within these high-level somatosensory areas do no longer operate in terms of receptive fields, but seem to have highly integrated representations of the sensory information which could represent steps towards object representation. However, the underlying network structure is poorly understood, and our knowledge of how this level of brain processing really works is highly limited.
h. 12-13
Martin Pearson, Bristol Robotics Laboratory, University of the West of England, Bristol, UK
Title: Understanding tactile sensory systems using a neurorobotic approach: the mammalian whisker sensory system
Abstract: Effective tactile sensing requires a tight coupling between sensors and motors, to understand this coupling we study a behaviour known as active touch. In humans this is exemplified by the way that we move our fingers across a surface to extract textural and fine spatial detail, or in the way that we manipulate objects to determine their gross structural composition. There are many other mammals that express similar active touch behaviour through their array of facial whiskers, or vibrissae (meaning vibrating). The rhythmic sweeping back and forth of the whiskers as such mammals explore their environment ("whisking") has attracted interest from those studying the processes of active touch for 3 important reasons; firstly, whisker arrays allow an exquisite degree of control over the application of stimuli; secondly, whiskered mammals are small enough to be easily handled and housed within the laboratory; and thirdly, there is extensive histological and physiological data available for the rat brain.
The presentation will summarise work that has contributed to this on-going study through the adoption of a neurorobotic approach. We focus here on the process of building an empirically based model of the mechanoreceptive sensory "front end" and primary afferent pathway from the rat vibrissae. We will also look at the issues of updating and maintaining such models with hard-real-time constraints using the finite processing and power resources of embedded programmable logic devices. Finally, a summary of more recent robotic hardware will be presented which have been used to study higher level neuroethological phenomena such as spatial navigation through touch and models of tactile sensor attention driving action selection.