Closed Levi-Armstrong closed 2 years ago
@marrts I made some slight modification to your changes. Thanks for the improvement.
Relevant PR when adaptive rho was disabled: https://github.com/tesseract-robotics/trajopt/pull/172
Look reasonable. Someone has run some tests and determined that this is a better default?
Yea both @marrts and myself have seen significant improvement in planning time especially for large problems.
Look reasonable. Someone has run some tests and determined that this is a better default?
Yes, this went from failing from iterations, which I found needed close to 10 million iterations to solving in about 10k iterations. Substantial speed improvement
The adaptive_rho was change because a unit test was failing here but now passes using the default setting indicating that either there was an issue in OSQP or TrajOpt that has been fixed. Have adaptive_rho enable has a signification impact on planning time so it is reverted to default value now that the setting is exposed.