New AI Agents Plan Cell Towers for Better Signal at Lower Cost

Placing cell towers is harder than it sounds. Radio waves bounce off buildings, get blocked by hills, and fade in odd patterns, so engineers usually walk the streets with equipment or run computer models that simulate every signal path. Both are slow and costly. A new research paper from Zihao Zhou, Zhaolin Wang, and Yuanwei Liu proposes letting AI agents do the planning instead.

The system starts with a 3D digital twin — a computer copy of the real neighborhood, complete with building shapes and heights. Two AI agents then work in a loop. The Placement Agent suggests candidate tower spots, using two modes: one that improves on promising ideas, and one that tries wilder options so it doesn't get stuck on a mediocre plan. The Reflection Agent then reads the simulation results and the map, figures out why coverage failed (a tall building blocking the signal, towers covering the same area twice, or a neighborhood with no signal at all), and turns those findings into advice for the next round.

In tests on two realistic urban scenarios, this approach beat older methods — rule-of-thumb formulas, machine-learning models, and even chatbot-style AI helpers. It also got close to the best possible coverage while running far fewer transmitter-level simulations, which are the expensive part. The researchers say the reflection step is what turns raw simulation data into useful guidance, and the two-mode setup keeps the search from settling for a bad answer.

The practical takeaway: network planning could get faster and cheaper, which matters as cities roll out 5G and prepare for 6G. Fewer wasted towers means lower costs for carriers — and potentially better, more reliable signal for you. The catch is that this is still a research paper tested on simulated cities, not a real deployment tool.