Trajectories Prediction of the Black-Tailed Gull Using the Inverse Reinforcement Learning

Abstract

Revealing the route selection of wild animals is of fundamental importance in understanding their movements and foraging strategy. In this study, we attached GPS loggers to black-tailed gulls Larus crassirostris and recorded their movement trajectories during their foraging trips. Using inverse reinforcement learning (IRL), we analyzed the factors that affected their route selection. During the training phase, using pre-defined feature maps, we estimated a reward map that may affect the decision making of black-tailed gulls. The reward map can be used for predicting the trajectories of the gulls during the test phase. In addition, the resultant weight vector enabled us to analyze to which degree the black-tailed gulls favor each area.