The dunnage dilemma: why the bin is still winning in automotive production

There is a category of manufacturing problem that rarely makes it onto a product roadmap or a capital investment deck, yet accounts for the majority of automation failures in automotive general assembly. It has no compelling acronym. It does not attract venture capital. It is not, by any reasonable measure, exciting. And yet, according to engineers who spend their working lives solving it, imprecise bin dunnage is responsible for roughly 80% of the disruptions encountered when OEMs attempt to automate line-side part picking.

That figure comes from Keith Sharkey, Account Manager at JR Automation, a Hitachi Group company, that has spent decades integrating robotic and vision-guided assembly systems across North American and European automotive facilities. The figure is clearly striking in its degree, and it points to something the industry has been slow to confront.

The discourse around automotive automation is largely dominated by the technologies themselves: collaborative robots, AI-powered vision, force-torque sensing, and ever-more sophisticated end of arm tooling (EOAT). What gets far less attention is the physical and logistical environment those systems must operate within - and in particular, the state in which parts arrive at the point of use.

"The people evaluating resource reductions on assembly lines don't fully realize how significant dunnage is," says Chris Buda, Engineering Manager at JR Automation, who has overseen automation programmes across both body shop and general assembly environments. "They assume parts arrive line-side in racks and that robots should simply pick them like operators do. In my experience, that's not at all realistic. And dunnage is a major issue. It's almost a disabler."

That word - disabler - deserves to sit with the reader for a moment. Not a complication. Not a nuisance. A “disabler”. A systemic condition that can render an otherwise well-engineered automation investment non-functional. Understanding why, requires tracing the problem back to its origins, which lie not on the assembly floor but considerably further upstream.

A problem born upstream

Automotive dunnage - the racks, trays, dividers, and foam inserts used to transport and present parts - was designed for an era when the end user of that system was a human operator. Operators are, as engineers will readily acknowledge, extraordinarily adaptable. A worker can assess a tilted part, compensate for a missing foam locator, reach into a corner of a rack, and make a safe pick in a fraction of a second. They bring judgment, proprioception, and a lifetime of learned physical intuition to every interaction. A robot brings none of those things.

When automation enters the picture, the mismatch between the dunnage ecosystem and the demands of robotic picking becomes immediately apparent. Robots require consistency. They need parts to be within a defined spatial envelope, oriented within acceptable tolerances, and accessible by the gripper geometry without obstruction.

And standard shipping dunnage, designed to protect parts in transit and maximize packing density, provides none of these guarantees.

The cost implications of addressing this at source - by requiring suppliers to use precision dunnage that places parts in repeatable positions - are substantial and cascading. A single precision rack can cost several thousand dollars more than its standard equivalent. Multiply that across the hundreds of racks in circulation at any given point in a production supply chain - simultaneously in use on the line, being loaded at a supplier, in transit, and being returned - and the capital investment becomes significant before a single robot has been specified.

There is also a density penalty. Precision locators and the spacing required for robotic gripper access mean fewer parts per rack. Fewer parts per rack means more racks per shipment. More racks per shipment means higher freight costs. The economics ripple upward through the supply chain in ways that are rarely fully accounted for when automation business cases are constructed.

"When labor wasn't a concern, this didn't matter," Sharkey observes. "Operators are very adaptable. But automation changes rack requirements completely."

The alternative - shipping densely in standard dunnage and repacking into precision process racks at the plant - transfers the cost rather than eliminating it. Labour is required at the point of decant, an activity that assembly plants are under pressure to remove rather than introduce.

"The alternative is repacking," Buda explains. "A supplier might ship parts densely, and then at the plant they're repacked into precision racks. That reduces the number of precision racks needed, but it increases labour costs at the plant - and assembly plants are trying to reduce exactly that kind of manual work."

The trade-off between precision dunnage costs and on-site repacking labour is a genuine dilemma, and there is no universal answer. The right solution depends on specific part geometry, the volume being run, plant space constraints, and the overall automation architecture. What is clear is that neither path is free, and the full cost of each is frequently underestimated at the point of investment decision.

Durability adds another layer of complexity. Dunnage is subject to forklift damage, rough handling, and the accumulated wear of a production environment. Organisations can be understandably reluctant to invest heavily in precision dunnage hardware that may be compromised before it delivers its intended value.

As part of its line-side integration work, JR Automation has deployed vision systems specifically to verify that adjustable dunnage is correctly configured and undamaged before a pick cycle begins - using vision not for the pick itself, but as a safeguard for the conditions that make reliable picking possible.

Why random picking remains an unsolved problem

If the economics of precision dunnage are challenging, the obvious question is whether automation systems can simply be engineered to handle imprecise bins - to pick from standard dunnage the way a human operator would. The answer illuminates how technically formidable the problem remains.

"True random picking has traditionally been impossible," says Buda. "Advances in vision have made semi-random picking more feasible - close to acceptable - but not fully there."

The challenge is not merely a vision problem, though vision is central to it. It is a convergence of multiple technical domains, each presenting its own set of unsolved problems. Vision systems must accurately characterize the position and orientation of every accessible part in a bin - a task complicated by variable lighting, part reflectivity, occlusion, and the dynamics of a moving production environment. If a part shifts between the moment a snapshot is taken and the moment the robot acts on that data, the pick fails.

"Consistency is critical," says Sharkey. "The system takes a snapshot, and everything must remain consistent after that. If a part moves after the image is captured, you need another snapshot adding to cycle time." At typical automotive cycle times - often in the range of 10 to 15 seconds - the margin for re-imaging is narrow. Parts positioned against the wall of a bin may be physically inaccessible regardless of how accurately they have been localized, a variable that no vision system, however sophisticated, can overcome without a redesigned approach to bin geometry or gripper reach.

Path planning must then translate the vision system's output into a collision-free trajectory for the robot arm and EOAT, accounting for bin walls, neighbouring parts, and the constraints of the gripper geometry. The gripper itself must be designed to handle the specific part reliably, which often means it cannot be a universal solution and adds yet another layer of engineering specificity to what superficially appears to be a commodity problem.

But the most underappreciated constraint may be part design itself. "This goes all the way back to vehicle design," says Buda. "Parts need to be designed for automation. You can't assume someone will figure it out later with a gripper. Assemblies already include features for manufacturing - locating holes, fixtures, key product characteristics - but what's missing is consideration for downstream automation at final assembly."

Body shop, where the economics of high-volume stamping have long driven the industrialisation of robotic handling, has developed a culture of designing parts with automation in mind - incorporating locating features, consistent datum points, and geometries that support reliable gripping. General assembly has been slower to develop this discipline. When product engineers are asked to incorporate features that support downstream automation, they face competing pressures: features add weight, may increase cost, and the benefit accrues to a manufacturing process that may be a decade away from being specified.

The organisational incentives do not always align with the systemic interest, and JR Automation's engineers encounter this misalignment regularly during early-stage programme discussions with OEM customers.

The buffer imperative

For automation integrators working in the space between the state in which parts actually arrive and the requirements of robotic assembly, the engineering challenge is as much about system architecture as it is about any single technology. One of the most consistent principles to emerge from JR Automation's programme experience is the necessity of buffering.

A line-side automation system that picks directly from incoming dunnage - with no intermediate staging - is a system with no tolerance for variability. The moment a rack needs to be changed, the line stops. "Without a buffer, the system is very fragile," says Buda. "Variability is unavoidable, whether from human or automated handling. Without a buffer, the first disruption will stop the line."

JR Automation's line-side sequencing systems typically incorporate staging as a structural element of the solution architecture - transferring parts from incoming dunnage into positions of known, repeatable presentation before the robot performs its assembly operation. This staging may take the form of fixed stands holding multiple parts in defined locations, or conveyor-based systems where parts are loaded onto pallets by operators and carried through pre-assembly operations before reaching the vehicle.

"We've implemented systems where operators load multiple parts onto a pallet, which moves along a conveyor," Sharkey explains. "A robot then performs assembly operations before the part reaches the vehicle. The buffer ensures parts are always available, supports option management, and provides time for rack changes without disrupting the line."

In either configuration, the buffer serves multiple functions simultaneously: it decouples the upstream variability of dunnage handling from the downstream precision requirements of robotic assembly; it supports option management by enabling multiple variants to be held ready; and it provides the dwell time needed to handle rack changes without stopping the line.

The staging approach does not eliminate the role of operators - it redefines it. Rather than performing repetitive pick-and-place operations at the point of assembly, operators become responsible for loading and managing the buffer. Whether this represents a meaningful reduction in labour depends heavily on the specific operation, but it does shift human activity toward tasks that are less ergonomically demanding and more amenable to job rotation.

What OEMs should be asking about AI Vision

The most active area of development in bin picking is the application of AI-powered vision - systems that use machine learning to identify and localize parts across a range of positions and orientations, reducing or eliminating the dependency on precision dunnage. The technology has advanced rapidly, and vendor demonstrations can be compelling. JR Automation, which has served as a testbed for advanced FANUC vision-guided robotics technology and worked at the frontier of semi-structured and structured bin picking for years, is closely tracking this development - and is candid about where the boundaries of reliable production performance currently lie.

"If AI can enable true random picking, it solves a major problem - eliminating precision dunnage requirements and reducing logistics costs," says Buda. "We've been chasing this for years. It's improved rapidly, but it's not fully there yet."

When evaluating AI vision claims, OEMs should move beyond cycle time and pick rate statistics and probe the operational conditions under which those figures were achieved. Vendor demonstrations frequently take place under controlled conditions - consistent lighting, clean part surfaces, a limited range of positions - that may not reflect the variability of a production environment across a full shift and multiple seasons.

Data architecture is another critical consideration, particularly for safety-related components. JR Automation's integrated vision systems, which incorporate 3D scanning, AI-based image processing, and camera-guided robotics, are designed with traceability as a core requirement rather than an afterthought. “AI systems are also evolving to create 3D models rather than relying solely on flat images - and OEMs need to understand how that data is stored, for how long, and how it can be retrieved,” says Sharkey.

OEMs with internal automation labs are well positioned to conduct their own validation testing rather than relying solely on vendor-provided data. Real-world validation - using actual production parts, actual dunnage, and actual lighting conditions - is the only reliable basis for a production deployment decision. "OEMs should conduct their own testing," Buda advises. "Many already do this in internal labs. Vendors demonstrate capability, but OEMs need to validate performance in real-world conditions."

The honest industry consensus is that AI vision has not yet fully solved the random bin picking problem, but that it is improving at a meaningful rate. When it does - when a system can reliably pick from truly random, unmodified shipping dunnage at production cycle times with acceptable error rates - it will eliminate the entire upstream cost structure of precision dunnage and fundamentally change the economics of line-side automation.

Closing "the last metre"

The phrase "the last metre" has become shorthand in automation circles for the gap between where the part arrives and where it needs to be - the final, often most problematic segment of the inbound logistics chain. It encompasses not just the physical pick, but the verification, sequencing, and error-proofing and traceability that surround it.

JR Automation has integrated barcode-based verification across numerous line-side sequencing programmes, using part identification as the foundation of a broader error-proofing architecture. But comprehensive verification requires consistent labelling across the entire supply base - a condition that is approached but rarely fully achieved. "Full error-proofing and traceability requires all parts to be labelled," is Sharkey's practical assessment.

Equally important is planning for the cases where errors occur despite best efforts. A robot that picks the wrong part creates a downstream problem that must be managed - the part cannot simply be discarded, and stopping the line has its own cost implications. "If the robot picks the wrong part, what happens next?" asks Buda. "You can't stop the line or discard it. Planning for those scenarios adds cost but provides significant value."

Designing fault-response protocols into the system architecture - what happens to a mis-picked part, how the error is logged, what the operator response procedure is - is a dimension of integration work that JR Automation's engineering teams build into programme scope from the outset. The value of that resilience extends well beyond production efficiency. Error-proofing is a significant quality lever. Cases where automation systems are bypassed - whether through system failure or operational pressure - have resulted in substantial rework events across the industry. "We've seen cases where systems were bypassed and hundreds of vehicles required rework," Sharkey notes. A system that prevents those bypass conditions, and that captures traceability data for every assembly operation, delivers quality value that is often underweighted in capital investment analyses focused primarily on labour reduction.

The parallel with body shop runs deeper than design culture alone. "In body shop, some OEMs moved assembly operations in-house to improve dimensional quality," Buda observes. "By controlling the process, they improved fit and finish. If quality is critical, bringing processes closer to final assembly makes sense." The principle applies equally to general assembly: tighter control over the conditions of part presentation, enabled by well-engineered line-side sequencing and buffering, produces quality outcomes that extend far beyond the direct labour saving on the bill of materials.

The integration imperative

What the dunnage challenge ultimately reveals is a gap in how automotive manufacturing approaches the integration of logistics and automation engineering. These disciplines have historically operated with limited coordination: logistics teams optimize for shipping density and rack cost; automation engineers optimize for pick reliability and cycle time; and the interface between them - the state in which a part arrives at the point of robotic use - falls into the gap between organisational responsibilities.

Closing that gap requires early and sustained collaboration between supply chain teams, dunnage designers, plant logistics, and automation engineers. It requires product designers to consider downstream automation requirements at the point of part design, rather than treating them as someone else's problem to solve later. And it requires organisations to account for the full system cost of dunnage decisions - not just the per-rack cost, but the downstream effects on freight, labour, automation complexity, and quality.

Space is a more significant constraint than is often acknowledged. Automation is frequently treated as an afterthought in plant layout planning, and the physical room required to implement effective staging and buffering is simply not available in many existing facilities. "Space is a major constraint. Automation is often an afterthought, and there isn't enough room," says Sharkey - an observation that JR Automation's integration teams encounter consistently, and that shapes the range of feasible solutions before a single robot has been selected.

The bin, in other words, is not merely a logistics problem. It is a systems problem - one that touches vehicle design, supply chain structure, plant architecture, and automation engineering simultaneously. Organisations that address it as such, and that engage automation partners early enough in the programme to influence dunnage design, buffer architecture, and error-proofing strategy, are the ones most likely to realize the full potential of the investments they are making.

JR Automation leverages decades of expertise in robotic integration, vision systems, and line-side automation to help automotive manufacturers tackle critical challenges like imprecise bin dunnage. By addressing these often-overlooked obstacles, JR Automation ensures seamless production, optimized efficiency, and maximum return on automation investments. Their comprehensive solutions are designed to transform the way manufacturers approach automation, from initial concept to full-scale implementation.