Comparison Between Deterministic and Deep Neural Network Based Real-Time Trajectory Prediction of an Autonomous Surface Vehicle

Document Type

Conference Proceeding

Publication Date



Computing Sciences and Computer Engineering


With the advancement of autonomous capability and the intelligence level of underwater and surface vehicles, cooperative operations with multiple underwater and surface autonomous vehicles have immense potential to study the ocean more efficiently. One of the challenges of cooperative operations includes keeping vehicles relatively close to each other but at a safe distance while avoiding interference between the acoustic communication links between underwater and surface vehicles and the instruments. Predicting the lead vehicle’s future trajectory can help avoid undesirable situations and keep communication infrequent to avoid interference. This paper compares a deep neural network-based trajectory planning model with three deterministic models: two-point, consecutive average, and linear regression. Several mission datasets were used to train the neural network from which future positions were predicted. One mission dataset was fed to all the deterministic models for prediction. The two-point model has the most accurate prediction among the deterministic models, while the consecutive average has the least accurate prediction. Overall, the deep neural network model has the most accurate predictions, though, in its current state, it might suffer from overfitting.

Publication Title