Learning Traffic Flow Dynamics with Stochastic Physics-Informed Neural Cellular Automata

TL;DR

This paper proposes a physics-informed neural cellular automaton for data-driven traffic flow modeling, which learns local transition rules while preserving physically meaningful constraints. The model is extended to stochastic dynamics and outperforms a standard neural cellular automaton. It successfully learns traffic dynamics from established models.

Why It Matters

Accurate traffic flow modeling is crucial for optimizing traffic light control, route planning, and autonomous vehicle navigation. This research enables more realistic and efficient traffic simulations, which can lead to reduced congestion and improved transportation systems. The approach can be applied to various complex systems beyond traffic flow.

Key Findings

• Physics-informed neural cellular automata can learn traffic dynamics
• Stochastic extension captures probabilistic transition rules
• Outperforms standard neural cellular automaton