system identification

[pierfabre]

To have an accurate simulation and for model-based control method such as NMPC we need to accurately model our system.

• model of the motor
• model of the pendulum

[pierfabre]

Model of the motor

Equations

with

Motor DataSheet

Initial guess based on datasheet

Blocked motor

which gives us

No load

which gives us

Experiment

• measured no load speed : 154 rad/sec in output of the encoder which is equivalent to 154 * 6.3 = 970 rad/sec
• measured no load : 0.13 A
• measured voltage : 23.9 V

which gives us ke=kc=0.235USI

We evaluate the response time Tr to be around 0.03s which gives us L=0.24 H

[pierfabre]

[pierfabre]

20240716 - Update on the System Identification of the Robot

• I used onshape-to-robot to generate a URDF from the CAD
• Then I ran a rollout simulation with this URDF and my simulator that uses pinocchio aba to compute the rigid body dynamics part
• I then used the experiment I performed in March on the real robot to perform system identification of the rotor inertia and of the friction coefs

This is the result obtained with the best set of parameters so far :

rollout.webm

This is a very nice result which gives me hope for the swing up using NMPC -> #86