Meeting Sasa&Nick&Gökhan&Iman - Githubissues

SasaKuruppuarachchi / adaptive_estimation

Codebse for the robotics project of ROBO.666

Apache License 2.0

2 stars 0 forks source link

Meeting Sasa&Nick&Gökhan&Iman #30

Closed EternalGoldenBraid closed 1 day ago

EternalGoldenBraid commented 3 days ago

Gökhans pointers

Choose a suitable level of researchy vs practical work.
Limit the scope from trying out everything to a few things.
- E.g. compare pros and cons of LQR and PID with tPC state estimation.
Implement in steps:
- Estimation without control
- Does the system behave as expected?
- Add control

General remarks

LQR we require a model of the system. I.e. $C$ and thus states must be sensible.
PID is more robust to model errors and we only care about observations.
Magnets for modifying physical system.
Don't worry about MPC for now.

SasaKuruppuarachchi commented 1 day ago

Regarding the uncertain tracking of the state

The system is optimising for too many outputs/parameters
Try fixing C matrix to I for example

Control Scheme

Compare performance with PID and LQR
With PID output of the state estimator does not need to be outputting the exact state
With the changing model (A) LQR will show interesting results
As the Second order information is produced in the estimator, it is good to send them to LQR
LQR is time invariant so this is good. (MPC is not as it is iterative)

Regarding the research

Limit the scenario to one line of the story to help publications within 6 months
Drop MPC for now