Warehouse Task Planning Optimization with NVIDIA cuOpt

Executive Summary

A global manufacturer partnered with SimpleRose to understand how intelligent planning and optimization could improve warehouse efficiency across capacity sizing, slotting, and daily task execution. SimpleRose developed a task-planning engine powered by NVIDIA cuOpt, designed for continuous re-optimization in a live environment. To validate its effectiveness, the POC used this engine to retrospectively re-plan historical activity by running a batch job that re-optimized each transaction based on a two-hour look-ahead window, improving both item placement and task sequencing.

During a one-month proof of concept, the system re-planned approximately 27,000 historical warehouse movement tasks and reduced total travel distance by 22%, revealing significant opportunities for labor savings and improved equipment utilization. Although congestion modeling was reserved for a later phase, the POC clearly demonstrated the value of GPU-accelerated optimization in real warehouse environments.

 

The Challenge: Improving Task Planning and Operational Flow

Warehouse operators moved items throughout the warehouse, using pallet jacks for lower locations and forklifts for higher racks. Planning for these activities relied on spreadsheet tools and simple rules that provided basic structure but not true optimization.

These methods did not minimize travel distance, balance operator workloads, or respond to real-time conditions. They also did not help operators choose efficient storage or pick locations, determine an effective task sequence, or identify opportunities to combine work and reduce unnecessary movement. Path selection was not optimized, which contributed to congestion and increased safety risks.

To assess how optimization could improve these decisions, the team conducted a proof of concept focused on location selection and task sequencing. Four weeks of historical warehouse movement data were re-planned to measure the impacts on travel, labor, and equipment use.

The Solution: Two Levels of Optimization

The proof of concept focused on improving two core decision layers: selecting efficient locations for put-away and pick tasks, and sequencing those tasks to reduce unnecessary travel. SimpleRose integrated NVIDIA cuOpt into its decision support system to create a planning engine capable of re-evaluating these decisions on a rolling horizon.

For the POC, the engine processed four weeks of historical warehouse movement data using a rolling two-hour look-ahead window. At each transaction, the model advanced this window and generated an updated plan. During each step, it evaluated alternative storage locations, considered future pick frequency, and identified opportunities to interleave tasks to reduce deadheading. The engine accounted for operator constraints, equipment capabilities, and zone layout to produce an optimized sequence of movements for every planning interval.

This rolling approach provided a realistic view of how the system could function in production, where task plans would be refreshed periodically throughout the day. By continually updating both location choices and task order, the engine demonstrated significant potential to streamline warehouse activity without requiring changes to existing workflows.

Impact and Results

The planning engine analyzed 27,424 tasks over the four-week period and produced updated plans at each transaction. The optimized sequence reduced total travel distance by 22% compared to the historical plan and improved equipment utilization by a similar margin. Average travel per task decreased from 365 to 285 meters, and estimated labor time per task fell from 12.06 minutes to 9.41 minutes.

Because the POC used a relatively short, two-hour look-ahead window, these results represent a conservative estimate of the system’s potential. The client confirmed that a longer planning horizon—on the order of eight to twelve hours—would be feasible in production because orders are released well in advance. With more visibility into upcoming work, the engine would have additional opportunities to select more efficient locations and create better task sequences, which is expected to increase the overall efficiency gain.

Future iterations will also incorporate routing and congestion management. While congestion modeling may offset some of the efficiency improvements from a longer planning horizon, the combined approach is expected to deliver performance well above the client’s target of a 15% reduction in travel.

Although the pilot focused on a single site, the findings indicated substantial opportunity across the broader warehouse network. At scale, the projected reduction in travel, balanced workloads, and improved equipment use could translate into significant annual savings in labor and operational efficiency.

Conclusion

The proof of concept showed that optimization can meaningfully improve how warehouse operations are planned and executed. By combining historical data with optimization-powered decision models, the team demonstrated a practical path to reducing travel, improving labor efficiency, and creating more adaptive task plans without changing existing infrastructure.

Beyond the operational improvements measured in the POC, the same models and data structures create a foundation for broader strategic and tactical planning. Industry research indicates that strategic slotting, improved layout design, and integrated returns handling can drive additional gains in throughput, accuracy, and cost efficiency. These opportunities were not addressed in the POC but represent meaningful extensions as the solution scales across the network.

Together, these operational and strategic capabilities position the organization to exceed its 15% improvement target and capture substantial long-term value across the warehouse portfolio.

SimpleRose helps organizations apply these methods at scale—solving complex planning and scheduling challenges across supply chain, logistics, workforce, and maintenance environments. By building tailored decision support systems and leveraging next-generation solvers, we enable teams to move beyond static planning and make better, faster decisions that deliver lasting impact.