Lessons from AlphaZero for Optimal, Model Predictive, and Adaptive Control
It provides a unifying framework for control theory, but appears incremental in extending known ideas to new domains.
The paper analyzes how AlphaZero's principles of approximation in value space and rollout can be applied broadly to optimal control problems, showing their integration with methods like model predictive and adaptive control.
In this paper we aim to provide analysis and insights (often based on visualization), which explain the beneficial effects of on-line decision making on top of off-line training. In particular, through a unifying abstract mathematical framework, we show that the principal AlphaZero/TD-Gammon ideas of approximation in value space and rollout apply very broadly to deterministic and stochastic optimal control problems, involving both discrete and continuous search spaces. Moreover, these ideas can be effectively integrated with other important methodologies such as model predictive control, adaptive control, decentralized control, discrete and Bayesian optimization, neural network-based value and policy approximations, and heuristic algorithms for discrete optimization.