A collection of Geometric Observers, Filters, and Optimization methods for Robotics & Computer Vision
A comparative benchmark of geometrically-aware sensor fusion architectures. Features a Dual Quaternion Geometric Observer (GeoDQ) using Screw Linear Interpolation (SCLERP), compared against Error-State (ESKF) and Unscented Kalman Filters (UKF-M) on the SE(3) manifold.
Highlights: 3.7x lower RMSE than ESKF, robust tracking at 7Hz updates, Python & JIT implementations.
View Project Page & Results View Code →Standard and JIT-optimized implementations of classical filtering approaches for 6DoF estimation. Includes Error-State Kalman Filter (Invariant EKF) and Manifold UKF. Used as baselines for the GeoDQ benchmark.
View Comparison View Code →TBD Algorithms for statistical analysis on manifolds. (Work in Progress)
Coming Soon