SiteSim

Warehouse robots can waste time queuing at crowded lifts while faster routes go unused. Integration is a broader industry hurdle: 42% of surveyed businesses cite it as their biggest warehouse-management-system challenge.(https://snapfulfil.com/news/snapfulfil-partners-with-manifest-as-the-official-warehouse-management-system) Self-Improving Warehouse Robots turns observed worker habits into tested robot operating rules. Our harness uses **warehouse simulation software with two floors, two freight lifts, and 20 robots** to evaluate changes across repeatable shifts. **MongoDB Atlas** stores each brain version and its evidence, while an ElevenLabs voice agent requests human approval for safety-sensitive changes. The first tested rule—take the stairs when another robot is waiting for the lift—reportedly eliminated lift-queue jams across 12 held-out simulated shifts** and improved the mezzanine score by **10%, without retraining the underlying model.