Robust Regret Control with Uncertainty-Dependent Baseline
By: Jietian Liu, Peter Seiler
Potential Business Impact:
Helps machines learn better with unknown problems.
This paper proposes a robust regret control framework in which the performance baseline adapts to the realization of system uncertainty. The plant is modeled as a discrete-time, uncertain linear time-invariant system with real-parametric uncertainty. The performance baseline is the optimal non-causal controller constructed with full knowledge of the disturbance and the specific realization of the uncertain plant. We show that a controller achieves robust additive regret relative to this baseline if and only if it satisfies a related, robust $H_\infty$ performance condition on a modified plant. One technical issue is that the modified plant can, in general, have a complicated nonlinear dependence on the uncertainty. We use a linear approximation step so that the robust additive regret condition can be recast as a standard $\mu$-synthesis problem. A numerical example is used to demonstrate the proposed approach.
Similar Papers
Distributionally Robust Regret Optimal Control Under Moment-Based Ambiguity Sets
Optimization and Control
Helps robots learn to control things better.
Revisiting Regret Benchmarks in Online Non-Stochastic Control
Optimization and Control
Teaches robots to learn better without knowing all rules.
Distributional Robustness in Output Feedback Regret-Optimal Control
Optimization and Control
Makes robots learn from mistakes better.