Currently the semi linear integrators are non-adaptive.
In principle, this implementation needs just deriving the right expression to the error calculation.
These papers are a good starting point.
It would be fantastic if adaptive step sizes appear in future releases.