Summary

Eurystic, a supply chain optimization and simulation consultant, worked with Pathmind on a simplified AnyLogic model showing deep reinforcement learning could help a crane stacking packages in a warehouse and increase its throughput.

 

Challenge

A warehouse operator needed to increase the throughput of cranes used to stack product packages on static racks within the warehouse. In order to achieve this goal, the intelligence about where to allocate a product to minimize material blocking and travel distance were crucial aspects.

Improving the operation of its warehouse cranes was aligned with a company-wide Digital Transformation initiative sponsored by the Operation Planning Director. Demand for products, production schedule and operating conditions in the warehouse varied a lot, which made it difficult to predict how the cranes should behave from one moment to the next.

The Pathmind policy showed a performance improvement of 300% over the random action baseline and achieved similar results to the original heuristic, but provided the possibility of drastically reducing the development time and the flexibility to replicate the solution in different warehouses.

Why Pathmind Was Needed

Eurystic has developed custom tailored heuristics for many warehouses under similar conditions. Although the results have been of high standard, the company was looking to challenge and go beyond the current solutions with new methodologies, in search of more flexibility and fast adaptation to different contexts.

Pathmind policies can be trained within simulations that include highly variable data, and they can learn to respond to that variability without needing to be rewritten. The cooperation between Eurystic, specialized in developing virtual environments to analyze processes and Pathmind, with its tool that leverages the power of AI, was a clear opportunity for synergy.

 

Outcomes

The Pathmind policy was trained in the simulation in a highly-variable context and learned to adjust the crane’s behavior to changes on the demand and the production schedule of the machines.

The Pathmind policy showed a performance improvement of 300% over the random action baseline and achieved similar results to the original heuristic, but provided the possibility of drastically reducing the development time and the flexibility to replicate the solution in different warehouses.

Eurystic distilled complex warehouse operations in a model that would demonstrate the value of reinforcement learning in tackling variable demands and operating conditions. The lessons derived from the single-crane simulation were applied to a three-crane warehouse scenario.

Start screen of overhead crane warehouse optimization model
Screenshot of warehouse optimization model running

To learn more about this warehouse optimization use case and details of the simulation model, contact us.

 

About Eurystic

Eurystic is a supply chain and simulation consulting company based in Argentina whose clients chiefly work in areas that include supply chain, logistics, network design and distribution.