Akreon Control Architecture

Akreon Control owns control and estimation models, formulations, and execution. This guide describes their source organization and data flow. For supported scope and future work, see the support roadmap.

Source organization

Paths below are relative to include/akreon/control/.

Path	Responsibility
control_traits.hpp	Time domains, model concepts, state/input/output types, and port roles
model/state_space.hpp	Continuous/discrete state-space representations
model/discretization.hpp	Zero-order-hold continuous-to-discrete conversion
model/plant_model.hpp	State-space models with typed input and output port layouts
feedback/	State-feedback synthesis, H2/H-infinity and polytopic formulations, and gain recovery
estimation/	Luenberger, H2/H-infinity, and polytopic observer formulations
control.hpp	Public umbrella header
bindings/ (repository root)	Optional C ABI and Python adapters

Public control APIs use control. Numerical types come from linalg, while optimization expressions, canonicalization, and solver results come from Convex.

Modeling and synthesis flow

1. Express a continuous or discrete model with StateSpace. Dimensions and scalar types are part of the matrix types.
2. Wrap it in PlantModel when synthesis requires role-tagged input/output ports, such as control, measurement, disturbance, or performance channels.
3. Select a feedback or observer formulation. The formulation assembles its optimization variables, objective, and constraints using Convex APIs.
4. Canonicalize and solve through Akreon Convex, inspecting its result and numerical status.
5. Recover controller or observer gains from the solved variables. Gain recovery belongs here because its interpretation depends on the control formulation.

Polytopic formulations describe synthesis across a collection of plant models. They share the modeling and solver layers rather than implementing a separate optimization stack. Supported families and their assumptions are documented in the individual feedback and estimation headers.

Dependency boundary

Akreon Linalg owns matrices, spaces, factorizations, and generic numerical kernels. Akreon Convex owns DCP composition, cones, canonicalization, and solver iteration. Control owns plant semantics, optimal and robust control formulations, estimation, and gain recovery. Future predictive-control, horizon-estimation, and barrier formulations belong here; their implementation status lives in the roadmap. There is no dependency on akreon-dynamics. Simulation results can be used to construct model matrices without introducing a library dependency in either direction.

API reference

Use Doxygen's namespace, class, and file indexes for model contracts, formulation assumptions, and gain-recovery algorithms documented in the headers.

Language boundary

Optional bindings adapt runtime arrays to the compile-time C++ APIs. The binding contract is maintained in bindings/README.md; build and notebook usage is in the repository README. The numerical algorithms stay in the C++ libraries.