Error Analysis in Motion Prediction for Distributed Interactive Applications

Although there are many prediction methods proposed to address the network latency problem, there is not much work conducted to study the accuracy of prediction algorithms. In this project, we propose to investigate the factors that affect the prediction accuracy of existing prediction algorithms for addressing the network latency problem of the demanding DIAs. We will study the characteristics of these prediction algorithms and analyze their accuracy profiles against various factors that affect the prediction accuracy. Based on these formal analyses, we will then develop mechanisms for estimating an error bound, which is composed of an upper error bound and a lower error bound, for each prediction made by the prediction algorithms. Some of the techniques that we will consider for estimating the error bound include the following:The statistical approach – by profiling the distribution of prediction errors during a sequence of predictions.The feedback approach – by referencing the errors made in recent predictions.The analytical approach – by analyzing the characteristics of the prediction model used by the predictor.Our investigation will include comparing the performance of the above techniques and the possibility of combining various approaches together. In addition to using a simulation environment, we will also study the performance of the new mechanisms for error bound estimation in a real application environment.