Embodied Computing: Self-adaptation in Bio-inspired Reconfigurable Architectures

This paper describes a bio-inspired architectural approach to design highly adaptive and reconfigurable systems in the context of mobile robotics. The aim is to design the hardware architecture of an intelligent controller for a robot that exhibits several behaviors such as landscape learning, obstacle avoidance, path planning, sensory-motor control. The concerned robot evolves in an indoor unknown environment and uses the visual informations coming from its input sensors to reach its goal. The Embodied Computing approach presented in this paper is employed in this context to integrate the reconfiguration management as a part of the behavior of the global system. Thus the controller will be able to self-organize its reconfigurable processing elements in order to adapt its architecture to the environment and to the robot actions. We propose in this paper a hardware implementation of the approach based on artificial neural networks. We present the results of our first prototype onto the last technology of FPGA.