报告人：澳大利亚国立大学 Yonhon Ng博士
报告概况：Bayesian filters are often used in localization and mapping tasks. They are commonly implemented on resource constrained systems for real time application. Extended Kalman filter (EKF) in particular has been widely used. However, it is known that EKF has inconsistency problem. A number of papers has studied the inconsistency issue and concluded that it arises due to rotation being seemingly observable. However, we show from our experiment that EKF can still produce inconsistent results even if rotation is given accurately. This suggests that other factors also contribute to the inconsistency observed. We propose a novel estimator that appropriately approximates the true measurement likelihood distribution. This is accomplished by fitting a degenerated Gaussian likelihood, which simplifies subsequent computation while ensuring consistent result. Monte Carlo simulation results for bearing-only localization show that our method compares favorably to popular non-linear filters in terms of accuracy, computational and memory requirement. The method is also generalizable to other problems such as distance-based localization.
报告人概况：Yonhon Ng is from Penang, Malaysia. He completed his Bachelor of Engineering (with First Class Honours and University Medal) in mechatronic systems from the Australian National University. He is currently a PhD student in the Research School of Engineering, Australian National University. His current research interests include Bayesian estimation, visual odometry, simultaneous localization and mapping (SLAM), optical flow and radio-based localization.