机器人RSS 20252025
SayCan 2: Grounding LLMs in Robot Affordances
Michael Ahn, Anthony Brohan, Noah Brown et al.Google DeepMind
摘要
SayCan 2 extends the original SayCan framework with improved affordance grounding and multi-step planning. We demonstrate robust long-horizon task execution on real robot platforms.
LLM groundingaffordancesmulti-step planninglong-horizon tasksreal robots
技术细节
数据集
SayCan DatasetOpen X-Embodiment
模型骨架
PaLM-E + Affordance Grounding Module
编码器
ViT + Language Encoder + Affordance Encoder
解码器
LLM Planner + Skill Selector
