Optimisation algorithms can reduce warehouse picking distance by 10%
Pickers in a busy warehouse can walk 15–20 kilometres a day. Significant distance savings can be obtained by carefully choosing the sequence the items are picked in, and in how large orders are split into individual pick-routes.
We ran a study to find out how much of that distance could be cut using modern optimisation algorithms, without changing the warehouse layout, staffing, or any physical infrastructure.
What the research tells us
- Routing optimisation alone is worth doing: ~3.9% distance reduction, improving about 1 in 3 routes
- Joint optimisation of splitting and routing is where the real opportunity lies: distance reductions of 10–25% depending on order complexity.
- Warehouse layout is a meaningful constraint on what is achievable
- With the right graph model, the approach is computationally viable for live warehouse systems
This research was conducted as part of the OWL project. At Solwr, we have built these findings into our WMS. If you want to explore what this could look like for your operation, reach out to our team.
The study in brief
- Data: one full week of historical pick-route data from a live warehouse (10–17 October 2023)
- Comparison: existing tag-based routing and rule-based splitting vs. optimisation-based routing and splitting using OR-Tools
- Metric: total distance walked
How the current system works
Splitting
Orders are typically split into pick-routes based on fixed rules:
- Carrier weight and volume limits
- Weight class (used as a proxy for pallet stability)
- Product temperature zones
- Rough location groupings
These rules are applied sequentially, and the resulting routes are then handed to the routing step. The two problems are never solved together.
Routing
Most WMS platforms route pickers using a tag-based policy: each warehouse location has a tag, and the picker visits locations in tag order. It is simple and consistent, but it does not account for the actual walking distance between locations. For example, a picker might need to walk an entire corridor to pick the next item in the sequence, rather than picking it later when they can take a shortcut and walk only half of said corridor.
What the optimised approach does differently
The new routing algorithm treats each pick-route as a variant of the classic traveling salesperson problem (TSP). It:
- Builds a distance matrix of actual walking distances between all pick locations
- Searches for the optimal visiting sequence within a time budget
- Respects weight-class constraints (heavier items picked before lighter ones)
- Falls back to the tag-based sequence if no improvement is found — so it never makes things worse
Splitting and routing solved jointly
- Proximity grouping is a byproduct of the global optimization, but the algorithm does not blindly enforce it, when it's detrimental
- Total distance across all routes for an order is minimised, not just each route individually
The computation challenge
Building a distance matrix for a large warehouse is computationally expensive. To solve the optimization problem, one needs to build a large matrix with hundreds of thousands of distances for a large order. This can take hours with standard distance computations.
The study used a reduced graph model — a simplified representation of the warehouse's corridor structure — which brings full matrix construction down to a few seconds. This makes real-time deployment feasible, with the caveat that corridor renaming requires a manual update to the model.
We are happy to talk through how your current picking setup compares and where the biggest gains are likely to be.