Revision history [back]

You can feed it those data sources, yes. The prediction step won't match your vehicle's kinematics, but you can try to compensate for that by having higher frequency odometry data and increasing the process_noise_covariance for the affected variables. The model doesn't work at the level of wheel velocities and doesn't use Ackermann kinematics. I don't intend on changing that model, but I do plan to move to a plugin-based architecture that will let users implement their own models. In the meantime, it would be pretty straightforward to add a different model with the existing architecture, though if you want to work with raw wheel velocities, you'd have to also add support for the reception of the those messages and new parameters (e.g., wheel radii, wheel separation distance, etc.).