Inspiration
The spark for this project came from realizing that warehouse efficiency isn't just about how fast robots move; it starts with the physical floor plan. Most layout tools focus on simple shapes, but real warehouses have obstacles, fixed aisle widths, and strict demand requirements. I wanted to build a "battle-tested" optimizer that could bridge the gap between abstract mathematical selection and real-world spatial constraints.
What it does
We successfully bridged the gap between theoretical math and physical reality by engineering a "fail-safe" feedback loop. By linking the ILP model and Placement Engine through iterative retry logic, the system guarantees a feasible, high-density layout even in irregular spaces. Seeing our multi-stage pipeline consistently outperform standard baselines in randomized stress tests validated that OptiSpace can handle the unpredictability of real-world logistics.
How we built it
I designed a multi-stage pipeline that faced different algorithms against one another: The Selection Layer: I used Integer Linear Programming (ILP) to determine the optimal count of each bay type based on cost and capacity. The Placement Layer: I integrated a shelf-packing engine to physically arrange those bays within the warehouse boundaries while avoiding obstacles. The Refinement Layer: I implemented a Simulated Annealing algorithm to refine initial ILP outputs, optimizing bay counts to achieve more cost-effective and high-density configurations The Stress Test: Engineered a comprehensive simulation suite to generate diverse warehouse dimensions and demand profiles, rigorously validating that the proposed algorithms consistently outperform baseline models
Challenges we ran into
The "Paper vs. Reality" Gap: Often, the ILP would suggest a mathematically perfect list of bays that simply wouldn't fit due to the warehouse's physical geometry. I solved this by implementing iterative retry logic: if the packer fails, the system automatically "tightens" the area constraints and re-solves the ILP until a feasible layout is found. Dynamic Aisle Modeling: Modeling aisles is a "chicken and egg" problem because the aisle area depends on the number of rows, which you don't know until you've already placed the bays. I developed an aisle-aware ILP variant that iteratively estimates row counts and converges on a realistic solution.
Accomplishments that we're proud of
We successfully bridged the gap between theoretical math and physical reality by engineering a "fail-safe" feedback loop. By linking the ILP model and Placement Engine through iterative retry logic, the system guarantees a feasible, high-density layout even in irregular spaces. Seeing our multi-stage pipeline consistently outperform standard baselines in randomized stress tests validated that OptiSpace can handle the unpredictability of real-world logistics.
What we learned
Through this project, I discovered that no single algorithm is a "silver bullet." While ILP provides mathematically exact solutions for selection, it can be too rigid for complex environments. I learned how to balance this by using Simulated Annealing and Greedy Multistart to find high-performance layouts in cluttered spaces where exact math struggles.
What's next for OptiSpace
We are evolving OptiSpace from a static planner into a dynamic warehouse twin. The next frontier is integrating real-time SKU velocity to enable automated seasonal re-slotting and 3D volumetric optimization for multi-level facilities. By shifting from reactive layouts to predictive spatial modeling, we aim to future-proof infrastructure before a single rack is ever installed.
Log in or sign up for Devpost to join the conversation.