AI/ ai · security · open-source · research

GraphChase Opens Up a Benchmark for AI Cop-and-Robber Games

A new open-source platform lets researchers test algorithms that model how law enforcement intercepts criminals across city road networks.

A research team has released GraphChase, an open-source platform for benchmarking AI approaches to urban pursuit problems.

Urban Network Security Games (UNSGs) model a specific real-world problem: how should law enforcement allocate limited resources across a city's road network to intercept a fleeing suspect? GraphChase gives researchers a common environment to build and compare algorithms for this class of multiplayer game, supporting both unweighted and weighted road networks — the latter meaning travel times vary by route, which is how actual cities work. The platform ships with learning-based baseline algorithms so new work has something to measure against.

The benchmark exposed a gap that matters beyond academia. Existing UNSG solvers degrade noticeably when road edges carry realistic, uneven travel costs rather than the uniform weights common in lab settings. That sim-to-real gap suggests algorithms that look competitive in controlled tests may not hold up when deployed against actual city topology — a relevant concern for any agency considering AI-assisted patrol routing or fugitive interdiction.

Most AI game-solving milestones — from chess to poker — involve two players and perfect or near-perfect information. Multiplayer games with asymmetric resources and messy real-world constraints are a harder and less-trodden category, and the absence of a shared testbed had slowed progress here the same way the lack of standardized datasets once slowed computer vision.

TR

The Revision

Written by an AI system from the public sources credited above. How we write →