Compare Ungar with OCS2

[edward9503]

Dear @fdevinc ,

Thanks for sharing such impressive work!

I also used the OCS2 before. Since you are in the same university but different groups, I was wondering have you ever compared the performance and computational speed with the OCS2? Is there any compelling reason why we should use Ungar instead OCS2? Many thanks!_

[fdevinc]

Dear @edward9503,

Thank you for your appreciation! Comparing OCS2 and Ungar:

• OCS2 boasts a longer history and numerous contributors on GitHub, whereas Ungar is a newer, individual effort. Ungar is currently evolving, with plans to improve interfaces, add solvers, and reduce compile times; upcoming updates will enhance installability and external library handling.
• OCS2 offers multiple solvers tailored for specific problem structures. In contrast, Ungar has only one that handles general sparse problems, albeit potentially slower. Ungar's optimizer is essentially the same as OCS2's SQP, but it uses the more versatile OSQP instead of HPIPM as a back end.
• Ungar prioritizes user-friendliness and runtime efficiency, especially for moderate-sized problems. We plan to share hardware demos in the coming months to provide more insights into Ungar's performance.

If you are accustomed to OCS2 and appreciate its interfaces, sticking with it is definitely reasonable! Unlike Ungar, OCS2 is more extensive, but it comes with a potentially steeper learning curve and demands more boilerplate code for implementation—exemplified by the available robotic examples.

Finally, there is always the option to use both at the same time: for example, you may use Ungar variable maps to prevent bugs when handling indices (see lines from 43 to 69 here). This is not the most comprehensive answer, but I hope it helps!