Scalable and Efficient Continual Learning from Demonstration via a Hypernetwork-generated Stable Dynamics Model
通过超网络生成的稳定动力学模型实现可扩展且高效的示范学习
机构 * Faculty of Electrical Engineering and Computer Science, Technical University of Berlin(电气工程与计算机科学系,柏林技术大学) ; Department of Computer Science, University of Innsbruck(计算机科学系,因斯布鲁克大学) ; Digital Science Center (DiSC), University of Innsbruck(数字科学中心(DiSC),因斯布鲁克大学) ; Department of Industrial Engineering, University of Trento(工业工程系,特伦托大学) ; Singular Research Center on Intelligent Systems (CiTIUS), University of Santiago de Compostela(智能系统研究中心(CiTIUS),圣地亚哥-德孔波斯特拉大学)
专题命中 机器人学习 :robotic(abstract);分类 cs.RO
AI总结 本文提出一种稳定的持续学习方法,利用超网络生成轨迹学习动力学模型和Lyapunov函数,提升稳定性与准确性,通过任务嵌入减少训练时间,实验证明其在多任务学习中优于现有方法。
Comments To appear in IEEE Transactions on Cognitive and Developmental Systems