This letter presents a novel unsupervised sensory matching learning technique for the development of an internal representation of three-dimensional information. The representation is invariant with respect to the sensory modalities involved. Acquisition of the internal representation is demonstrated with a neural network model of a sensorimotor system of a simple model creature, consisting of a tactile-sensitive body and a multiple-degrees-of-freedom arm with proprioceptive sensitivity. Acquisition of the 3D representation as well as a distributed representation of the body scheme, occurs through sensorimotor interactions (i.e., the sensory-motor experience of the creature). Convergence of the learning is demonstrated through computer simulations for the model creature with a 7-DoF arm and a spherical body covered by 20 tactile fields.

In recent years, advances and improvements in engineering and robotics have been strengthening interactions between biological science and robotics in the goal of mimicking the complexity of biological systems. In this paper, motor control paradigms inspired by human mechanisms of sensory-motor coordination are applied to a biologically-inspired, purpose-designed robotic platform. The goal was to define and implement a multi-network architecture and to demonstrate that progressive learning of object grasping and manipulation can greatly increase performance of a robotic system in terms of adaptability, flexibility, growing competences and generalization, while preserving the robustness of traditional control. The paper presents the neural approach to sensory-motor coordination and shows preliminary results of the integration with the robotic system by means of simulation tests and experimental trials.

In this letter we describe a hippocampo-cortical model of spatial processing and navigation based on a cascade of increasingly complex associative processes that are also relevant for other hippocampal functions such as episodic memory. Associative learning of different types and the related pattern encoding-recognition take place at three successive levels: (1) an object location level, which computes the landmarks from merged multimodal sensory inputs in the parahippocampal cortices; (2) a subject location level, which computes place fields by combination of local views and movement-related information in the entorhinal cortex; and (3) a spatiotemporal level, which computes place transitions from contiguous place fields in the CA3-CA1 region, which form building blocks for learning temporospatial sequences. At the cell population level, superficial entorhinal place cells encode spatial, context-independent maps as landscapes of activity; populations of transition cells in the CA3-CA1 region encode context-dependent maps as sequences of transitions, which form graphs in prefrontal-parietal cortices. The model was tested on a robot moving in a real environment; these tests produced results that could help to interpret biological data. Two different goal-oriented navigation strategies were displayed depending on the type of map used by the system. Thanks to its multilevel, multimodal integration and behavioral implementation, the model suggests functional interpretations for largely unaccounted structural differences between hippocampo-cortical systems. Further, spatiotemporal information, a common denominator shared by several brain structures, could serve as a cognitive processing frame and a functional link, for example, during spatial navigation and episodic memory, as suggested by the applications of the model to other domains, temporal sequence learning and imitation in particular.

For gradient descent learning to yield connectivity consistent with real biological networks, the simulated neurons would have to include more realistic intrinsic properties such as frequency adaptation. However, gradient descent learning cannot be used straightforwardly with adapting rate-model neurons because the derivative of the activation function depends on the activation history. The objectives of this study were to (1) develop a simple computational approach to reproduce mathematical gradient descent and (2) use this computational approach to provide supervised learning in a network formed of rate-model neurons that exhibit frequency adaptation. The results of mathematical gradient descent were used as a reference in evaluating the performance of the computational approach. For this comparison, standard (nonadapting) rate-model neurons were used for both approaches. The only difference was the gradient calculation: the mathematical approach used the derivative at a point in weight space, while the computational approach used the slope for a step change in weight space. Theoretically, the results of the computational approach should match those of the mathematical approach, as the step size is reduced but floating-point accuracy formed a lower limit to usable step sizes. A systematic search for an optimal step size yielded a computational approach that faithfully reproduced the results of mathematical gradient descent. The computational approach was then used for supervised learning of both connection weights and intrinsic properties of rate-model neurons to convert a tonic input into a phasic-tonic output pattern. Learning produced biologically realistic connectivity that essentially used a monosynaptic connection from the tonic input neuron to an output neuron with strong frequency adaptation as compared to a complex network when using nonadapting neurons. Thus, more biologically realistic connectivity was achieved by implementing rate-model neurons with more realistic intrinsic properties. Our computational approach could be applied to learning of other neuron properties.

A neural network model of intrahippocampal and hippocampo-cortico-gangliobasal loops allows the robotic implementation of the spatial information contained within cognitive maps in neural space, into temporo-spatial sequences of movements during goal-oriented navigation in outer space. At the representation level, the intrahippocampal loop features place cells in entorhinal cortex and transition cells in CA3-CA1 which constitute, as a population, spatial and contextual maps, respectively. Path integration converges with place cell information in the subiculum. The spatial representation in deep and superficial layers of the entorhinal cortex are dissociated; the unidirectional connections between these two layers close the intrahippocamal loop. The prefrontal cortex, at the junction between representation and implementation, receives from three hippocampal subsystems and stores a global graph-map of an environment and the goal locations on this map. The diffusion of activation from active goals through the graph allows path selection in the motivational limbic prefrontal cortex and planning in the executive lateral part. The top-down output from prefrontal cortex and the bottom-up output from the hippocampus combine onto the accumbens the first stage for the stepwise selection and implementation of the optimal actions in the direction of the goal. Proactive and reactive functioning modes are dissociated.

The aim of the work is the integration of an anthropomorphic robotic platform, starting from a neurophysiological model of grasping, in order to provide tools for an experimental validation of the model and also to provide new anthropomorphic solutions for robot grasping. The resulting robotic system is composed of an anthropomorphic arm/hand system and visual and tactile sensors. Grasp planning, control and learning have been achieved by a neural approach, inspired by a model of the inter-connections among the brain areas involved in grasping, as formulated by neurophysiologists. After a description of objectives and specifications for the robotic platform, the system is illustrated and experimental results are reported. Finally, the results are discussed as a starting point for current activities, involving the development of novel human-like robotic components and the implementation of more sophisticated learning schemes.

Visually guided landmark navigation is based on space coding by hippocampal place cells (PC). A biologically realistic architecture of cooperative-competitive associative networks (implemented as a control system for mobile agents) emulates PC activity during local navigation in exploration and goal-retrieval. The system builds and stores panoramic views from landmarks and compares them with current inputs. Mismatch-induced low levels of recognition trigger a vigilance burst, which favors either the recognition of an alternative place category or the creation of a new category. The sole implementation of visual "What" and "Where" information does not restrain the generality of the model since several modalities could cooperate to give rise to more robust place field spatial categories. Providing the system with real visual inputs automatically extracted from a natural environment demonstrates that interspecies differences in PC coding result more from characteristics of the visual systems than from differences in processing. Conversely, differences in PC multiple codes within a system result from different levels of processing and/or different degrees of multimodality. Each code could be used within different navigational strategies. A control system derived from the model allows a mobile agent to learn a few places and the associated actions required to reach a goal. Generalisation property of the model provides the capacity to join the goal from any place in the learned environment.

A two-layered network model which learns and performs two-joint arm reaching movements in 3-D space and is consistent with neurobiological data on cell activities recorded in the motor cortex during these movements is proposed. These experimental results show that: (1) neurons which command muscle synergies are broadly tuned to preferred directions in the 3-D space, (2) the cell's preferred direction rotates with the initial position of the arm, and (3) the rotation of the arm in space and the rotation of the whole population of preferred directions are highly correlated. The motor cortex is thus working in an arm centered coordinate system, and the available visual information about the desired trajectory should be translated in this coordinate system for the appropriate command of the arm. The authors show how this computation can be performed by processing units which model the properties of the cortical column as the basic neuronal circuits common to all cortical areas