[2M5-OS-19c] 世界モデルと知能(3/4)

[2M5-OS-19c-03] Inference and Dynamic Optimization of Task Goals by Collaborative Robots Using Deep Learning

〇Shun Hiramatsu¹, Shingo Murata¹ (1. Keio university)

Keywords:deep learining, robot learning, human–robot collaboration, goal directed action, prediction error minimization

Collaborative robots are expected to generate actions by inferring a task goal from an environmental situation while optimizing the inferred goal based on human partner's behavior. The objective of this study is to develop a computational framework that enables collaborative robots to learn such an ability. For this, we propose a framework consisting of three deep neural networks. A goal inference network is jointly trained with a goal recognition network to infer a latent goal only from the initial image of a task space. An action generator generates the prediction about a visuomotor state from its current state and the inferred latent goal. During action generation, the inferred latent goal is optimized to minimize visual prediction errors for adapting to human partner's goal. To evaluate this framework, we conducted an experiment on a collaborative object arrangement task. Experimental results demonstrate that the robot with the framework realized a successful collaboration.