Design and implementation of a sensorimotor network for chemical sensing using a mobile robot platform

Many current robotic systems are application-specific and have difficulty if the environment changes. These controllers do not scale well with increased task complexity, and rely on widely used high quality sensors. However, biological systems exhibit impressive adaptability. Therefore, self-organizing architectures should be incorporated into robotic systems to allow for emergent intelligence and robustness given less than optimal sensors and environments. In this study, a flat, fully-connected sensorimotor architecture was implemented on the EvBot III platform for the application of chemical sensing. The network was trained to associate increased alcohol concentration with increased battery charge. Seven training and testing experiments were conducted using different learning protocols. Although the sensorimotor network was shown to be a good initial step towards robotic reflex behavior, the robot was unable to successfully learn to home to the alcohol source.