Spatial computing in structured spiking neural networks with a robotic embodiment

One of the challenges of modern neuroscience is creating a "living computer" based on neural networks grown in vitro. Such an artificial device is supposed to perform neurocomputational tasks and interact with the environment when embodied in a robot. Recent studies have identified the most critical challenge, the search for a neural network architecture to implement associative learning. This work proposes a model of modular architecture with spiking neural networks connected by unidirectional couplings. We show that the model enables training a neuro-robot according to Pavlovian conditioning. The robot's performance in obstacle avoidance depends on the ratio of the weights in inter-network couplings. We show that besides STDP, critical factors for successful learning are synaptic and neuronal competitions. We use the recently discovered shortest path rule to implement the synaptic competition. This method is ready for experimental testing. Strong inhibitory couplings implement the neuronal competition in the subnetwork responsible for the unconditional response. Empirical testing of this approach requires a technique for growing neural networks with a given ratio of excitatory and inhibitory neurons not available yet. An alternative is building a hybrid system with in vitro neural networks coupled through hardware memristive connections.