A Regret Minimization Approach to Iterative Learning Control
This work addresses iterative learning control for systems with uncertain dynamics, offering a more robust approach, though it appears incremental as it builds on recent non-stochastic control advances.
The paper tackles the problem of model-based policy learning under uncertain, time-varying dynamics by proposing a new performance metric called planning regret, which uses worst-case regret instead of standard stochastic assumptions, and designs an iterative algorithm that shows improved robustness to model mismatch and uncertainty, with empirical evidence of outperforming existing methods on benchmarks.
We consider the setting of iterative learning control, or model-based policy learning in the presence of uncertain, time-varying dynamics. In this setting, we propose a new performance metric, planning regret, which replaces the standard stochastic uncertainty assumptions with worst case regret. Based on recent advances in non-stochastic control, we design a new iterative algorithm for minimizing planning regret that is more robust to model mismatch and uncertainty. We provide theoretical and empirical evidence that the proposed algorithm outperforms existing methods on several benchmarks.