This paper introduces two locally optimal feedback controllers designed for real-time implementation of constrained optimal control. Unlike conventional approaches that require extensive computation or rely on sub-optimal approximations, the proposed controllers can be computed efficiently online while handling nonlinear dynamics, nonlinear costs, and box-constrained control inputs. Their performance is validated through theoretical analysis, simulation, and robot experiments.
Why it matters: Real-time robot control often requires balancing accuracy, efficiency, and robustness under constraints. These controllers make constrained optimal feedback control practical for real-world robots, improving their ability to handle perturbations and modeling errors with low computational cost.