| Robots need to act in the real world, but they are constrained by weight, power, and computation capability. Artificial intelligence (AI) techniques try to mimic the clever processing of living creatures, but they lack embodiment and a realistic environment. This thesis introduces a novel robotic architecture that provides slender robots with massive processing and parallel computation potential. This platform allows the investigation and development of AI models (the brain) in interaction with its body and environment. Our robotic system distributes the processing on three biologically correlated layers: the Body, Brainstem, and Cortex; on board the robot, on a local PC, and on a remote parallel supercomputer, respectively. On each layer we have implemented a series of intelligent functions with different computational complexity, including a binaural sound localization technique, a bimodal speech recognition approach using artificial neural networks, and the simulation of biologically realistic spiking neural networks for bimodal speech perception. |
