Distributed Diffusion-Based Manifold Particle Filters for Orientation Estimation

Determining orientation is a fundamental problem in fields such as robotics, computer vision, and autonomous driving. In three-dimensional space, orientations can be represented in many forms, including unit quaternions or points of the Special Orthogonal Group (SO(3)). These representations present challenges due to their non-Euclidean nature. This paper addresses the problem of computing a joint Bayesian estimate of a rotation given the observations of a set of receivers that consist of corrupted and transformed versions of the true orientation. We evaluate the performance of the described algorithms via numerical simulations in terms of accuracy and communication requirements.