Foundations of Artificial IntelligenceCycle I · 2025-2
A hybrid search agent: ant colony algorithms and Dijkstra for rescue in maze-like environments
An agent for search-and-rescue in unknown, maze-like terrain. Ant algorithms handle the exploration; once the victim is found, Dijkstra plans the route out.
Search-and-rescue operations in unknown or collapsed environments are a hard case for artificial intelligence: the terrain is maze-like, uncertain and dynamic, and response time is critical. Drones and autonomous robots suit these scenarios precisely because they can operate where a human presence would be dangerous.
Following the approach of Husain et al. (2022), this project builds an agent that splits the problem in two and uses a different family of algorithms for each half. Ant algorithms drive the exploration phase, where collective swarm behaviour covers unknown ground efficiently. Once the victim is located, Dijkstra takes over and computes the optimal rescue route — a deterministic answer to a problem that no longer needs stochastic search.
The result is a balance between efficient exploration and optimal path planning, with a model of cooperation between distributed agents that explore, communicate and reorganize into a relay network between the base and the victim. Experiments show a significant reduction in search time and better environment coverage compared with purely heuristic agents.