Sharing the Control Authority Between Deep Reinforcement Learning and Model Predictive Control: Application to Multi-Class Transportation Networks

What the math gives to ML

The paper offers a transferable multi-rate control mechanism: a slow model-based predictive controller computes coarse, constraint-aware decisions, while a fast deep policy supplies high-frequency actions. The neural-network abstraction is a hierarchical residual controller in which the policy does not need to solve the full constrained optimization online. The key engineering variable is the time-scale ratio between MPC and neural updates, creating a measurable compute-performance crossover. The paper does not provide a general stability theorem, so transfer should be tested through constraint violations, action-spectrum separation, and performance as the update ratio changes.