【报告摘要】The talk will briefly introduce the outcomes of asymptotically solving several non-convex localization and pose estimation problems using two-step estimators. This approach involves: 1) initially generating a globally consistent estimator of the true parameters; 2) utilizing the estimate obtained in Step 1 as the starting point for a one-step Gauss-Newton iteration. The merit of the two-step estimator lies in: 1) its asymptotic optimality, provided the globally consistent estimate derived in Step 1 is O_p(1/sqrt{m}), where m represents the data size; 2) its linear scaling with m, making it computationally efficient. Thus, the key challenge lies in achieving a globally consistent estimate with a rate of O_p(1/sqrt{m}) using the available data. We accomplish this for several localization and pose estimation problems through techniques such as model transformation, bias analysis, and bias-eliminated least squares estimation.
【报告人简介】牟必强，中国科学院数学与系统科学研究院副研究员，于2008年从四川大学获得工学学士学位，2013年从中国科学院数学与系统科学研究院获得理学博士学位，研究兴趣包括系统辨识、机器学习、定位和位姿估计等。
