In a world of constraints, perfection is not always the goal.
Fuel is finite. Computation is bounded. Time is short.
And so, the system does something wise—it stops chasing the best, and starts aiming for the sufficient.
This is the art of suboptimal control.
At first, it sounds like a compromise. And in some ways, it is. Suboptimal control does not seek the mathematically ideal path, the global minimum of cost, or the exact theoretical solution. Instead, it asks a more grounded question:
What control action will give me good performance, fast, reliably, and within the limits of what I have?
In systems like autonomous aircraft, suboptimal control is not failure.
It’s strategy.
Imagine an onboard computer with limited processing power, flying through wind, under time pressure. Solving a full optimal control problem (like in infinite-horizon LQR or nonlinear MPC) might be too slow, too complex, or too fragile. The system can’t afford it.
So instead, we choose an approach that gets close—close enough to stabilize, to track, to conserve energy, to survive—without burning cycles chasing the unattainable.
Suboptimal control strategies include:
– Heuristic controllers based on tuned rules or approximations.
– Greedy algorithms that choose locally best actions without long-term optimization.
– Reduced-horizon MPC, where only a few steps ahead are considered, or constraints are relaxed for tractability.
– Approximate dynamic programming, where exact cost-to-go functions are estimated but not fully solved.
Each method is shaped not by ambition, but by context.
In aircraft control, suboptimal strategies can:
– Keep path tracking accurate under high wind, even when full prediction is impossible.
– Provide good-enough regulation under input saturation.
– Deliver fast decision-making on embedded processors during critical maneuvers.
The brilliance of suboptimal control is that it admits something most optimal frameworks ignore:
The world is imperfect.
And in that world, the best controller isn’t the most precise—it’s the one that works, under pressure, with limits, in real time.
This doesn’t mean we stop aiming high.
It means we define “high” in terms of real mission goals: safety, efficiency, reliability—not theoretical elegance.
Because intelligence is not only knowing what is best—
It is knowing what is enough, and when enough is everything you need.