[2C6-OS-7c] OS-7 (3)

[2C6-OS-7c-01] Data-driven modeling of trajectory in human chase and escape behaviors

〇Kazushi Tsutsui¹, Keisuke Fujii^1,2, Kazuya Takeda¹ (1. Nagoya University, 2. RIKEN)

Keywords:Trajectory prediction, chase and escape

Chase and escape behaviors are fundamental skills in many sports and are crucial for the survival of many animals in the wild. The prediction of the locomotion trajectory of the opponent is important for achieving the purpose (i.e., escape or interception) but is difficult because a number of factors, such as strategy, kinematic ability, and surroundings, are involved in making the decisions on locomotion trajectory. In particular, the modeling of escape behavior has not been successful due to its diversity. Here, we challenged to predict the locomotion trajectory of evader and pursuer using a data-driven model based on a recurrent neural network (RNN). Our results showed the superior performances of the RNN model to the conventional models. These results suggest that there are some rules for escape and chase behaviors, and that it is possible to predict the trajectory by learning from repeated observations.