[2G1-GS-2d] 機械学習：シーケンシャルデータの処理

[2G1-GS-2d-03] Finding an Equilibrium Trellis in Dynamics

〇Naoto Iwahashi¹ (1. Okayama Prefectural University)

Keywords:Trajectory Inference

It is important to infer multiple trajectories in a wide range of applications, such as human--robot physical interactions and autonomous driving. This is because it is often difficult to infer exactly one trajectory mainly owing to the lack of sufficient observed information and the effect of bifurcation. This study addresses the inference of equilibrium trellises that include multiple local optimal trajectories with given dynamics and objectives. The proposed method is shown to determine informative trellises, which are difficult for even humans to find.