Intelligent Sensor Uncertainty Modeling Techniques and Data Fusion
This paper presents a novel strategy to develop a sensor model based on a probabilistic approach that would accurately provide information about individual sensor's uncertainties and limitations. The strategy also establishes the dependence of sensor's uncertainties on some of the environmental parameters or parameters of any feature extraction algorithm used in estimation based on sensor's outputs. To establish this dependence, the approach makes use of a neural network (NN) that is trained via an innovative technique that obtains training signal from a maximum likelihood (ML) estimator. The proposed technique was applied for modelling stereo-vision sensors and an infrared (IR) proximity sensor used in the robotic workcell available in the Robotics and Manufacturing Automation (RAMA) Laboratory at Duke University. In addition, the paper presents an innovative method to fuse the probabilistic information obtained from these sensors based on Bayesian formalism in an occupancy grid framework to obtain a three-dimensional model of the robotic workspace. The capability of the proposed technique in accurately obtaining three-dimensional occupancy profile and efficiently reducing individual sensor uncertainties was validated and compared with other methods via experiments carried out in the RAMA Laboratory.