AI and the World's Hard Problems · Theoretical computer science · arXiv · 2026-10-22
Adaptive Grid Refinement for Optimal Feedback Control
The work addresses solving the Hamilton-Jacobi-Bellman PDE to obtain optimal feedback control, where fixed grid methods can fail to capture intricate details of the solution. A reinforcement learning-based algorithm dynamically refines the grid, building higher node density in regions with large discretization error while reducing the node population. This targets improved solution accuracy while optimizing computational resource usage.