Sens-Perceptor: Image based evidence formation module as a logical sensor for robot hand preshaping

The authors attempt to develop and implement a system executing a type of action according to a decision based imperfectly sensed and processed input data from multiple sensors. The system consists of three main modules, namely the “Sens-Perceptor” module (SP), the Fractal Inference Network (FIN), and the Neuro Controller (NC). Inputs to the Sens-Perceptor are the images sensed in an environment. In an implementation, these images are those of objects contained in the robot hand workspace. The objects considered are imperfectly known graspable parts available to the robot hand in its environment. Since the objects are imperfectly known, a fractal based methodology is developed for assigning initial beliefs at the low level processing of object images. The FIN is used for determining the task oriented hand preshapes of a multifinger robot hand for ill-perceived object contours. The neurocontroller locally activates the FIN by firing network nodes that bear high relevance to the output of the Sens-Perceptor module