Typical SymForce Integration Workflow

[mikoff]

Hey SymForce community!

I am curious about typical usage patterns and best practices for integrating SymForce into existing codebases.

Here's the scenario I have in mind:

• Define symbolic models (states, residuals) using SymForce, placing them in a dedicated folder (e.g., models).
• Generate C++ or Python code from these symbolic definitions, storing the autogenerated states and residuals code separately (e.g., generated).
• Implement interfaces to external optimization libraries. (Or is SymForce’s built-in optimizer stable and mature enough for direct use?)
• Clearly isolate autogenerated code from manually written code for better maintainability.

Does this scheme looks legitimate? Are there any other alternative or preferred workflows? Trying to find some real usage examples on github is not an easy task - all the links point back to the main SymForce repo :)

Any insights or references are greatly appreciated!
Thanks a lot in advance!

[aaron-skydio]

Thanks, also curious what others in the community are doing.

As far as what we do at Skydio, generally we follow the pattern you're describing, yes. And we do use SymForce's built-in optimizer for lots of problems, in production, on our drones and elsewhere. I'd say it's stable, yes, if it covers the features you need for your problem. It doesn't claim to implement everything you might ever want, so absolutely if you need something that isn't implemented it's totally valid to use SymForce-generated functions with another optimizer.