GINIO: A Geometric SO(3)-Equivariant Interface for Neural Inertial Odometry

GINIO presents a geometric SO(3)-equivariant interface for neural inertial odometry that addresses arbitrary IMU mounting conventions. The method ensures neural network measurements transform consistently by enforcing that the mean transforms as a vector and covariance transforms congruently as a second-order tensor under SO(3) rotations. This approach integrates learned measurements into filtering pipelines while maintaining proper geometric constraints for rigid body motion estimation.