Abstract
Data detailed on Robotics have been pr esented. According to news reporting originating in Nanjing, People's Republic o f China, by NewsRx editors, the research stated, "Pick-and-place manipulation is a pivotal component in many robotic tasks. In unstructured, dynamic and uncerta in environments, controllers are required to learn and generalize pick-and-place manipulation from multitype demonstrations, which are often represented in Euc lidean (e.g., position), and non-Euclidean (e.g., quaternion, symmetric positive definite (SPD) matrix) data." Funders for this research include Natural Science Foundation of Jiangsu Province , National Natural Science Foundation of China (NSFC), Zhejiang Lab. The news reporters obtained a quote from the research from Southeast University, "While many learning from demonstration (LfD) methods excel in manipulation tas ks with single-type data, the adaptation of motion skills across different data types remains largely open. In this paper, we propose a two-stream Sequence-to-S equence Network (S2SNet) that utilizes multiple tangent spaces to learn motion s kills from multi-type demonstrations. The S2SNet is capable of adapting the lear ned motion skills to different types of waypoints, such as position waypoints, S PD-based waypoints, and quaternion waypoints. Additionally, we introduce a demon stration segmentation operation within S2SNet to enhance its generalization abil ity across different types of waypoints and to reduce the dependency on a large volume of real demonstrations. Experimental evaluations show that S2SNet outperf orms state-of-the-art methods in passing through different types of waypoints."
NETL
NSTL