The authors prove a population L2 stability estimate and finite-sample certificate for one policy-evaluation step in a neural HJB solver with learned dynamics, plus multi-step propagation through greedy improvement, with experiments on high-dimensional control tasks.
On the convergence of policy iteration in stationary dynamic programming.Mathematics of Operations Research, 4(1):60–69, 1979
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Stabilized neural Hamilton--Jacobi--Bellman solvers: Error analysis and applications in model-based reinforcement learning
The authors prove a population L2 stability estimate and finite-sample certificate for one policy-evaluation step in a neural HJB solver with learned dynamics, plus multi-step propagation through greedy improvement, with experiments on high-dimensional control tasks.