画像キャプショニングと大規模言語モデルによるロボット動作列生成モデリング

抄録

In this paper, we describe a method for generating robot action sequences using a large-scale model, which allows achieving appropriate robot behaviors while reducing the learning costs. We utilize fine-tuned large-scale models to build the action generation modeling. The task for validating action generation involves placing cubes into drawers. Action sequences are generated based on the current state to achieve the goal, and new action sequences are generated in case of action failure or external interference.