Akreon Tutorial Series

From Dynamics to Feedback with Uncertainty

A Cartpole-based tutorial arc for understanding decision systems under uncertainty.

One system, the full decision loop

This tutorial arc uses the cartpole system as a compact example of Akreon's broader decision engine thesis. Across the series, we move from nonlinear dynamics to state-space models, feedback synthesis, observer design, optimal control, and uncertainty modeling.

The goal is not just to stabilize a toy system. The goal is to expose the recurring structure behind real decision problems: state, dynamics, objectives, constraints, uncertainty, and feedback.

Technical reference

The consolidated notebook contains the nonlinear model, state-space derivation, controller and observer synthesis, numerical results, diagnostics, and closed-loop simulations for the complete arc.

Read the rendered notebook

The notebook is the detailed technical reference. The chapter summaries below provide a guided path through its equations, code, plots, and results.

Watch the tutorial arc

The six-part companion series walks through the derivations, design choices, results, and limitations developed in the notebook.

Watch the YouTube playlist

Six parts

  1. Part 01

    Equations of Motion

    Build the coupled nonlinear cartpole dynamics from cart translation, pole rotation, actuator force, and gravity. This establishes the physical model and identifies the nonlinear terms that shape the control problem.

  2. Part 02

    Linearization to State Space

    Linearize the dynamics around the unstable upright equilibrium, assemble the continuous-time state-space model, and examine its poles, controllability matrix, and controllability Gramian.

  3. Part 03

    State Feedback via LMIs

    Convert the bilinear Lyapunov stability condition into a convex LMI using a change of variables. Recover the feedback gain, close the loop, and verify that the resulting poles are strictly stable.

  4. Part 04

    Observer Feedback

    Use position and angle measurements to reconstruct the full state with a Luenberger observer. Observability diagnostics and dual LMI synthesis lead to an observer-based dynamic compensator.

  5. Part 05

    Optimal Control Design

    Compare LQR and LQG designs with H₂ state-feedback and observer synthesis. The chapter connects cost weighting, disturbance channels, estimator design, and packed dynamic compensators.

  6. Part 06

    Uncertainty Modeling

    Represent pendulum-mass variation with a polytopic model and synthesize gains that satisfy every vertex. Solver diagnostics and nonlinear rollouts compare nominal and robust controllers under disturbance.