symforce

Why we included this project

SymForce goes after a pain that shows up in almost every real robotics codebase: handwritten Jacobians and hand-tuned kernels drift out of sync with the math they are supposed to model. You define the problem symbolically using geometric and camera types with Lie-group support, and SymForce generates branchless C++ (or another target language) with the tangent-space Jacobians computed for you. The bundled factor-graph optimizer runs those generated functions in real time, so the formulation you prototype in Python is the same one that runs in production. Teams doing calibration, bundle adjustment, SLAM, or nonlinear MPC on autonomous vehicles and drones get a practical way to cut boilerplate and keep the math and runtime code consistent.