Evolutionary calibration of sensors using genetic programming on evolvable hardware

In order to retain some degree of decisionmaking ability in a complex and dynamic environment, there are many attempts to build autonomous mobile robots. However, conventional methods pay little attention to the unreliability of sensors. Because of the corruption by noise and the difference of sensitivity, even the same kind of sensors shows different observation under the same condition. This causes a problem that a minor change of the environment of the sensor system has great influence on the perception ability of the robot. To improve the reliability of sensors, we present a method for evolutionary calibrating sensors using genetic programming as calibration mechanism. In our approach, sensor calibration logic is implemented on evolvable hardware. Therefore, as the learning goes on, sensor interpretation circuit reconfigures itself to a more suitable form during runtime. Through two experiments on different tasks, we confirmed that our method improved the correctness of interpretation significantly.