Runtime & deployment

kontrolem_controllers is the runtime consumer of the Kontrol’Em design pipeline: chainable ros2_control controller plugins that load linear state-space controllers (LQR, LQG, H∞) exported by state_space_setup_assistant and run them on a robot. No synthesis happens at runtime — just realtime-safe matrix–vector math on gains computed offline.

Design-time (Kontrol’Em, Python)	Runtime (this repo, C++)
URDF → LTI plant → Riccati / H∞	load <name>_ros2_control.yaml
runs once, offline	runs at the controller rate, on-robot

The bundle carries the operating point and the gains; the controllers rebuild the deviation state x = [q − q_eq, q̇], apply the law, and write effort commands. See the export contract for the exact schema and conventions — it is the single source of truth shared by the exporter and these loaders.

Packages

Package	Description
kontrolem_controllers	umbrella metapackage
kontrolem_controllers_base	shared library: artifact loader, static-gain & dynamic-compensator cores (Tustin discretization), validity guard, chainable controller base
lqr_controller	static state feedback u = u_eq − K·x
lqg_controller	LQG output-feedback compensator (Kalman + LQR as one LTI), runs from position-only measurements
hinf_controller	H∞ output-feedback compensator (hinf / hinf_mixsyn designs)

Contents

• Getting started
  • Build
  • Run the LQR demo (mock hardware)
  • How a controller finds its artifact
  • Standalone vs chained
  • Validity guard
  • Tests
• Export contract — <name>_ros2_control.yaml
  • The file
  • State, input, and output conventions
  • Control laws
  • Discretization (runtime responsibility)
  • Validity region
  • Loader requirements
  • Referencing an artifact at runtime
• Controllers
  • lqr_controller
  • lqg_controller
  • hinf_controller
  • Shared architecture
  • Chaining
• Gazebo Fortress physics validation
  • Architecture
  • Run it
  • Test
  • Notes / gotchas
• Isaac Sim physics validation
  • Architecture
  • Run it (two terminals)