Self-Organizing Control: From Autonomous Robots to Intelligent Prostheses

We consider biologically inspired robotic agents that are controlled by a neural network with fast synaptic plasticity. The dynamics of the synaptic strengths arises from the interaction of the robot with its environment and serves to establish an efficient sensorimotor loop. We show how explorative behaviors are formed by self-organizing control from a tabula-rasa initialization. The resulting behaviors are adapted to both the robotic hardware and the environment and arise from a compromise between two antagonistic objectives, namely sensitivity to external signals and predictability. The self-organization paradigm is applied to the control of a myoelectric hand prosthesis with the goal of an automated, user-specific interaction between patient and prosthetic device.