USA-Net: 机器人记忆的统一语义和可负担表示（英文版）

英文标题：USA-Net: Unified Semantic and Affordance Representations for Robot Memory中文摘要：本研究提出了一种简单的方法来构造一个可微分的世界表示，该表示将场景的语意和空间可负载特征编码成一个映射表，从而实现了同时优化场景语意和可负载特征的端到端可微规划。英文摘要：In order for robots to follow open-ended instructions like "go open the browncabinet over the sink", they require an understanding of both the scenegeometry and the semantics of their environment. Robotic systems often handlethese through separate pipelines, sometimes using very different representationspaces, which can be subop