RIANN—A Robust Neural Network Outperforms Attitude Estimation Filters

Inertial-sensor-based attitude estimation is a crucial technology in various applications, from human motion tracking to autonomous aerial and ground vehicles. Application scenarios differ characteristics of the performed motion, presence disturbances, environmental conditions. Since state-of-the-art estimators do not generalize well over these characteristics, their parameters must be tuned for individual circumstances. We propose RIANN, ready-to-use, neural network-based, parameter-free, real-time-capable inertial estimator, which generalizes across different dynamics, environments, sampling rates, without need application-specific adaptations. gather six publicly available datasets we exploit two method development training, use four evaluation trained estimator three test with varying practical relevance. Results show that RIANN outperforms filters sense it much better variety motions conditions sensor hardware frequencies. This true even if are on each dataset, whereas was completely separate data has never seen any datasets. can applied directly adaptations or training therefore expected enable plug-and-play solutions numerous especially when accuracy but no ground-truth tuning disturbance uncertain. made available.