A computationally efficient method for online joint state inference anddynamical model learning is presented . The dynamical model combines an a prioriknown state-space model with a radial basis function expansion representingunknown system dynamics . The method uses an extended Kalman filterapproach to jointly estimate the state of the system and learn the systemproperties . The model is inherently adaptive and can learnunknown and changing system dynamics on-the-fly . By usingcompact radial basis functions and an approximate Kalman gain, thecomputational complexity is considerably reduced compared to similarapproaches . The approximation works well when the system dynamics exhibitlimited correlation between points well separated in the state- space domain . It is exemplified via two intelligent vehicle applications where it isshown to have essentially identical system dynamics estimation performance and (ii) be real-
Author(s) : Anton Kullberg, Isaac Skog, Gustaf HendebyLinks : PDF - Abstract
Code :
https://github.com/nhynes/abc
Keywords : state - system - dynamics - model - space -
- Spark in R using sparklyr
- Kaggle Learning
- fast.ai Machine Learning
- Mathematics for Machine Learning by Marc Peter Deisenroth, A. Aldo Faisal and Cheng Soon Ong
- Computational Linear Algebra by Rachel Thomas
- Machine Learning Nanodegree
- Deeplearning.ai Specialization by Andrew Ng
- Mathematics for Machine Learning Specialization