How FM Logistic tackled the traveling salesman problem at warehouse scale with AlphaEvolve

FM Logistic used Google Cloud's AlphaEvolve to optimize warehouse routing by evolving a better traveling salesman algorithm, achieving a 10.4% efficiency gain over an already-optimized baseline.

• •AlphaEvolve is an evolutionary coding agent powered by Gemini models that iteratively generates, scores, and refines algorithms rather than using fixed rules.
• •The team provided their existing routing algorithm as a "seed" program, and AlphaEvolve mutated it against a custom evaluation function using 60 real warehouse tours.
• •The evaluation penalized constraint violations such as exceeding forklift capacity, duplicate assignments, FIFO violations, and excessive computation time.
• •The winning algorithm introduced three improvements: density-based anchor selection for tour starts, two-step filtering with distance simulation, and flexible route building that returns unfit orders to the pool.