L4 exists to correct a habit most control engineering builds without meaning to: the belief that a controller tuned to work well at several points must work everywhere in between. It usually doesn’t, and this level is built around making that failure visible before it’s fixed properly.
A deliberately staged failure
The signature experiment in L4 constructs a transition-instability case by design: several individually stable, well-tuned operating points, interpolated in the way engineers have always interpolated gains, producing an unstable trajectory in between. Watching your own correctly-tuned gains fail this way is what makes the LPV alternative land — a formal, LMI-backed guarantee replaces an interpolation you can no longer fully trust.
Offline design, online execution
This level makes the offline-design-versus-online-execution boundary explicit for the first time: the LMI synthesis runs offline in MATLAB before the lab session; what actually executes on STM32 in real time is nothing more than gain interpolation against the scheduling parameter. That boundary — what’s guaranteed offline, what merely executes online — is the same one every embedded control engineer has to internalise, and L5 depends on it directly.