Physics-informed neural networks often learn wrong solutions for some control problems
Low Residuals, Wrong Value: Learning the Viscosity Solution of Hamilton-Jacobi-Isaacs Equations with PINNs
Machine Learning
Summary
Some computer methods use tiny errors in equations to find solutions, but for certain math problems called Hamilton-Jacobi-Isaacs equations, small errors do not guarantee the right answer. The authors show that standard neural networks can learn wrong solutions that look good by error measures but are actually far from the true solution. They test different neural network setups and find some work better by changing how information about geometry is included. They also propose a way to check if a solution is really correct without knowing the exact answer beforehand.
What this means in practice
- •For control system engineers: Improve neural network methods to reliably find correct value functions in pursuit-evasion and related differential games.
- •For robotics navigation teams: Ensure learned solution quality in path-planning algorithms by using network parameterizations that incorporate geodesic distances.
Authors
Raul Jimenez, Pablo Tejerina-Pérez, Leonid Sarieddine, Pedro Tarancón-Álvarez, Marisol Traforetti, Pau Solé-Vilaró, Oleksii Ignatenko, Nazar Tkach, Pavlos Protopapas
Abstract
Physics-informed neural networks (PINNs) and other residual-trained solvers judge a candidate solution by the size of its sampled PDE residual. For first-order Hamilton-Jacobi equations this is not sufficient: the same equation and boundary data admit Lipschitz almost-everywhere solutions that differ from the viscosity solution by $O(1)$ while violating the equation only on sets of measure zero. We study whether residual minimization selects such solutions in a fixed-horizon pursuit-evasion game whose Hamilton-Jacobi-Isaacs value function is known in closed form on the circle $S^1$ and on the flat tori $\mathbb{T}^{2}$ and $\mathbb{T}^{4}$. We train four neural parameterizations with ten seeds on each geometry, 120 runs in total. A standard network with periodic Fourier features converges to folded almost-everywhere solutions in all 30 runs: small residuals and the expected capture-layer gradient, yet values that are wrong over most of the domain. A network with periodicity imposed through the loss recovers the viscosity solution in one and two dimensions but becomes unreliable in four. Two parameterizations that receive the geodesic distance do substantially better: a single network fed the wrapped displacement and its distance recovers the viscosity solution in all 30 runs, while a minimum over periodic image charts succeeds in 27 of 30 and introduces a wrong-image failure mode. Finally, for the convex Hamiltonian considered here, we use the semiconcavity of the viscosity solution to build a mesh-free finite-difference test that needs only forward network evaluations; it detects the folded solutions and the harmful min-of-charts failures without access to the exact value function. A small sampled residual thus does not by itself validate a learned Hamilton-Jacobi solution; the parameterization and independent structural diagnostics can be essential for selecting the viscosity solution.