Improvements for Continuous Koopman

EthanJamesLew / AutoKoopman

AutoKoopman - automated Koopman operator methods for data-driven dynamical systems analysis and control.

GNU General Public License v3.0

53 stars 10 forks source link

Improvements for Continuous Koopman #82

Closed EthanJamesLew closed 3 months ago

EthanJamesLew commented 6 months ago

From meeting 02/23

[ ] Reset at every step
[ ] Simulate using just the $A$ and $B$ matrices
[ ] Adaptive step

Look into the accuracy here--the fit is suspect.

This is needed for the reachability work.

CC @Abdu-Hekal

Abdu-Hekal commented 5 months ago

@EthanJamesLew Continuous Koopman works well for systems with no inputs but seems to suffer for systems with inputs. For example in the autokoopman-introduction notebook and for the PendulumWithInput benchmark. These two figures show the difference between discrete and continuous learning, with the worse fit for continuous learning. Could this potentially be due to a mismatch in input step and system step which isn't handled well by continuous learning?

EthanJamesLew commented 3 months ago

closing for now