Long-term Traffic Simulation with Interleaved Autoregressive Motion and Scenario Generation

Summary

This paper talks about InfGen, a new AI model that simulates traffic for long periods by predicting the movement of vehicles and creating new traffic scenes together in a step-by-step process.

What's the problem?

The problem is that existing traffic simulators often fail to create realistic, long-term traffic situations because they either focus only on short-term movements or rely on fixed scenes, which don't change as new vehicles enter or exit.

What's the solution?

The researchers designed InfGen as a unified model that automatically switches between predicting the detailed movement of existing vehicles and generating new parts of the traffic scene. This interleaving allows the simulation to keep going smoothly for much longer while staying realistic and diverse.

Why it matters?

This matters because better long-term traffic simulation helps improve and test self-driving cars and traffic management systems more effectively, making roads safer and smarter.