Эксперимент: Multiplicative Manifold Unscented Recurrent Cell (#296)

{ "worked": true, "confidence": 8, "verdict": "Built quaternion exponential/logarithm operations, multiplicative sigma-point retraction, iterative quaternion averaging, tangent covariance estimation, and a first-order control. The manifold method preserved unit quaternion norms to 2.22e-16 and reduced propagated-mean error from 0.04178 to 0.00180 on average for sigma <= 0.4. At sigma=0.8 it degraded, so the benefit is real but limited to moderate uncertainty.", "metrics": { "baseline": "First-order/EKF-style propagation: mean error 0.04178 for sigma <= 0.4 and 0.27927 at sigma=0.8; additive quaternion norm errors were 0.00374, 0.01477, and 0.05744 at sigma=0.1, 0.2, and 0.4.", "idea": "Multiplicative manifold UT: maximum quaternion norm error 2.22e-16; propagated mean error 0.00180 for sigma <= 0.4 and 0.29570 at sigma=0.8. UT covariance traces tracked Monte Carlo traces through sigma=0.4." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 manifold_ukf_experiment.py", "files": [ "manifold_ukf_experiment.py", "results.json", "run.log" ], "limitations": "No trainable GRU/MLP, sequence forecasting, learned process noise, NEES evaluation, GPU experiment, or runtime/FLOP comparison was tested. The experiment uses a synthetic quadratic tangent transition and a fixed UT configuration." }