Robust state estimation is critical for enabling reliable autonomous robot operations in challenging environments. To estimate the state, heterogeneous sensor fusion is commonly employed to enhance the reliability against perceptual failure. However, most known methods for sensor-fusion are brittle to dynamic perceptual condition changes due to the use of hand-tuned and time-constant error models. This paper introduces ROSE, a Robust Online-adaptive State Estimator, capable of adapting uncertainty statistics for individual multi-modal estimates in real-time to perform reliable sensor-fusion for robot state estimation. The proposed method leverages theory from adaptive Kalman filtering and extends it to optimization-based methods, to improve estimation accuracy while enabling integration of delayed heterogeneous sensor inputs. ROSE has been thoroughly evaluated by simulation studies and real-world demonstrations using a high-speed off-road vehicle navigating complex unstructured terrains and performing aggressive motions.
ROSE: Robust State Estimation via Online Covariance Adaption
Springer Proceedings in Advanced Robotics
The International Symposium of Robotics Research ; 2022 ; Geneva, Switzerland September 25, 2022 - September 30, 2022
08.03.2023
16 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Batch Measurement Error Covariance Estimation for Robust Localization
British Library Conference Proceedings | 2018
|Online Adaption einer Software Funktion zur Schwingungsunterdrueckung
Kraftfahrwesen | 2009
|Online-Adaption einer Software-Funktion zur Schwingungsunterdrückung
Tema Archiv | 2009
|Online covariance estimation for novelty‐based visual obstacle detection
British Library Online Contents | 2017
|