It’s not you: Pallet detection is harder than expected

Pallet detection seems straightforward: detect a standardized object with known dimensions. 

ifm made this assumption when the development of the Pick and Drop System (PDS) began seven years ago. Tens of millions of successful picks later, it became obvious why, and how, everyone underestimates the scope of this challenge when deciding to build a proprietary solution.

Developing a successful pallet detection system is certainly achievable. However, approaching it without understanding all the factors leads to wasted resources, delayed deliverables, missed opportunities, and plenty of frustration before creating a viable product.

Eventually, these problems affect engineering managers and the C-suite, not just developers. 

ifm can help. Having already encountered and solved the most difficult challenges, the ifm PDS solution offers an efficient and affordable pallet detection system.

When “Simple” isn’t simple

The complexity behind building a pallet detection system reveals itself gradually. 

At the start of deployment, the algorithm successfully detects standard B pallets at the first facility. The project appears to be progressing on the right track. Then, the complexity of an unstructured environment sets in.

CHEP pallets at the second facility require dataset updates.

Shrink-wrapped pallets go undetected.

Lighting changes affect recognition. 

Rackable plastic pallets, bottomless A pallets, and modified designs with stringers all demand system adjustments. 

Even familiar CHEP pallets, when damaged, require new algorithm updates.

What started as a simple detection problem became an endless cycle of modifications. Each update is simple enough, but new anomalies keep surfacing. 

Now, the challenge comes into focus. It's not that pallet detection is inherently difficult, but it's handling an infinite combination of anomalies in an unpredictable world that quickly overwhelms development capacity. 

The real problem: Variability at scale

Detecting every possible pallet, in every possible condition, requires managing virtually infinite variability at an industrial scale. The system must simultaneously handle:

• Multiple pallet types
• Unexpected designs
• Damaged units
• Diverse load configurations

The real goals: Efficiency and availability

The more work an autonomous mobile robot or automated guided vehicle performs, the more valuable it is to end users. Robust pallet detection improves three key metrics: 

• Speed: Can it reduce 30 to 60 seconds from current mission times?
• Accuracy: Does it still miss pallets, drop pallets, or pick bad pallets?
• Reliability: Will it work in any condition and adapt to updates quickly? 

Key insights from seven years of pallet detection development

ifm had a unique opportunity. The development team leveraged decades of automation experience with the resources to spend years refining the algorithm to solve for every detected anomaly and edge case.

Along with producing a valuable solution for the AMR market, the process also yielded a road map of what pallet detection developers must navigate:

• Every facility is bespoke. The system must work with each location’s unique variables and expectations, not just their pallets.
• Edge cases are the norm. The system must accommodate new pallet types, modified units, and varying load types. In some locations, wrappings vary seasonally. 
• Maintenance is perpetual. Developers should expect ongoing development well past the first launch. The algorithm will require regular testing, system updates, and validation across different facilities.
• The resource investment is substantial. Plan to create and maintain a dedicated algorithm team. Allot continuous time and budget for ongoing deployment support at each facility.

None of these challenges are insurmountable. Most development teams are capable of creating a proprietary pallet detection algorithm. However, it is important to consider that dedicating resources to this goal will take capacity from other projects. 

Each company will need to answer: What is the opportunity cost of that development time?

Keep reading: Deciding to make or buy a pallet detection system