Collision avoidance based on predictive probability using Kalman filter

In this paper, a collision avoidance algorithm based on predictive probability using the Kalman filter is proposed. Existing algorithms are based on the Time to the Closest Point of Approach (TCPA) and Distance to the Closest Point of Approach (DCPA) derived from geometric data between own ship and target ships using current states (heading angle and velocity). When compared to these approaches, the proposed algorithm is capable of predicting the states and easy to plan optimal path by considering the uncertainty in avoidance situation based on the number of ships. In this study, an Unscented Kalman Filter (UKF) is used due to the nonlinearities in the state variables to be predicted. By using the predicted state variables of the ships calculated via the UKF's prediction step, the positions of the number of ships after a certain time are derived as a predictive probability. Then, the validation of the proposed algorithm is examined by planning the optimal path in the collision avoidance simulations by applying the predictive probability.