ICMLJun, 2023

LIV: 语言 - 图像表示和奖励在机器人控制中的应用

TL;DRLanguage-Image Value learning (LIV) is used to pre-train a multi-modal representation for vision-language and reward learning, which can perform robotic control and reward specification in different environments using unseen robots or humans, and consistently outperforms the best prior input state representations and reward specification methods for policy synthesis.