Decision-Prediction Sensor Fusion for Intelligent Mobile Device Navigation

Most of the modern intelligent mobile devices such as intelligent vehicles or robots rely on sensor fusion to perceive the environment and make the decision on direction by traditional maximum likelihood (ML) criterion and possible direct decision feedback. To optimally fuse the sensor observation, we propose a novel approach called decision-prediction fusion (DP fusion). It further includes the previous decision as well as the previous state in the state transition concept of Kalman filter to derive the a prior probability of the current state. Thus traditional sensor ML fusion is converted to maximum a posteriori probability (MAP) detection by this approach. In this paper, we investigate service/rescue robot navigation problem to illustrate DP fusion theory and its application. The robot fuses sensors' observation to decide the direction of its destination. To derive the a prior probability for DP fusion, we establish the relationship of the current direction of destination with the previous decision: the angle of destination's current direction is nearly the same as the angle between the previous decision on direction and the true direction. Combining with sensor observation model, we formulate the sensor fusion problem as a prediction problem in the form of state space model. Then we derive DP fusion algorithm based on MAP detection. Simulations show that the proposed DP fusion outperforms the traditional scheme and is more robust to parameter variations such as observation SNR.