机器人NeurIPS 20252025
RoboDreamer: Learning Robot Tasks from Textual Descriptions
Kevin Black, Noah Brown, Sergey Levine, Pieter AbbeelUC Berkeley
摘要
We present RoboDreamer, a system that enables robots to learn new tasks from textual descriptions alone. RoboDreamer uses a language model to decompose task descriptions into executable sub-goals and leverages pre-trained skills to accomplish each sub-goal without additional demonstrations.
text-to-tasklanguage-guidedsub-goal decompositionpre-trained skillszero-shot learning
