System fitting problem(System states prediction for single quadcopter)

dmpcwkh commented 3 months ago

Why is the loss value relatively small during preprocessing, but becomes large during block processing? In preprocessing, the first 246 steps, min loss model, loss:0.0002497693107229894 Finish loading the pretrained model Batch 1, first 0 steps Saved min loss model, loss:5723460041.991171 I sincerely hope you can help me answer

wenjianhao commented 3 months ago

Hello,

Are you running codes on the same dynamics example or a different one?

dmpcwkh commented 2 months ago

I'm very sorry that I just saw your message. I was busy with my mentor's project some time ago, which caused related research to stall. When I am reproducing your experiment on the quadcopter drone, I would like to ask you some questions: (1) Do you use the torque and lift of the drone as input or the motor speed as input (2) How should I set the input when obtaining the drone status through the drone model? I use the motor speed as input, but when the motor speed is entered as a random value within a certain range, its status will behave strangely. How should I solve this problem and collect data? This question has been confusing me, please help me solve it

wenjianhao / Deep-Koopman-learning-of-nonlinear-time-varying-systems

System fitting problem(System states prediction for single quadcopter) #2