Gradient based training of r for EKF (compare to current noise estimator). Use softplus reparameterization.
Gradient based training of (q, r, alpha) for EKF and Lofi.
I. Greenberg, S. Mannor, and N. Yannay, “The Fragility of Noise Estimation in Kalman Filter: Optimization Can Handle Model-Misspecification,” arXiv [cs.LG], Apr. 06, 2021 [Online]. Available: http://arxiv.org/abs/2104.02372
I. Greenberg, S. Mannor, and N. Yannay, “The Fragility of Noise Estimation in Kalman Filter: Optimization Can Handle Model-Misspecification,” arXiv [cs.LG], Apr. 06, 2021 [Online]. Available: http://arxiv.org/abs/2104.02372