Management Summary
Mobile picking robots are increasingly marketed as a flexible and scalable automation solution for warehouses. However, a closer analysis reveals that they often fail to address the core inefficiencies of manual picking processes. Instead of eliminating the non-value-adding task of walking, as in goods-to-person (GtP) systems, mobile robots automate that very task—at lower speed and higher cost.
The key problem: mobile picking robots replace expensive human labor with expensive, limited-capability machines, without improving the ratio of productive to unproductive movements. They add complexity, require ongoing maintenance and fleet orchestration, and still depend on human intervention for error handling and replenishment. While their visual appeal and modularity make them attractive for pilots or innovation showcases, they rarely deliver significant gains in throughput or ROI in full-scale operations.
There are niche applications where mobile picking robots can add value—particularly in brownfield warehouses, labor-constrained environments, or temporary setups. But for most operations, especially those seeking long-term efficiency, they represent a technological detour rather than a strategic solution.
Decision-makers should scrutinize automation proposals not based on novelty, but based on structural process improvement.
The Initial Doubt
During some team management meeting back in 2016, our VP said that we were given the task of evaluating the addition of mobile picking robots to our corporate product portfolio. Most of my peers were quick to conclude that mobile picking robots would be “nice” or “good to have”. In my mind, however, I felt significant resistance against the idea. Initially, I could not quite articulate why I felt mobile picking robots are not something we should have. This bothered me, so I sat down in an effort to hyper-caffeinate myself and put my thoughts on paper. This worked.
Next day, in the continuation of the meeting, I ask my peers and my boss: “Why do we automate?” The replies revolved around cost savings and productivity and were generally sensible, though a little vague. I wanted to express the reasoning differently: “We automate because in warehouse operations, there are very costly assets – humans – spending much of their time performing non-value adding tasks – walking around. We want to improve the ratio of productive to unproductive movements of this expensive asset.” This, in a nutshell, is why there are goods-to-person picking systems: we eliminate the most time-consuming non-value adding task (walking around) and instead have storage totes do the travel task for us.
So far, so good.
The Fundamental Problem
Let’s take that very explanation and build on it to explain why mobile picking robots do not make an awful lot of economic sense: “We automate because we want to improve the ratio of productive to non-productive movements of our most expensive assets. With mobile picking robots, however, we merely replace one expensive asset with another expensive (and far less capable) asset performing the same poor ratio of productive to unproductive movements.” Accordingly, we won’t win anything. And because the new asset is so much less capable – much slower and much less versatile – we probably end up spending more AND we will still need humans to complement the robots.
On Paper, It Moves. In Practice, It Crawls
A mobile picking robot, in its most common form, navigates somewhat autonomously through warehouse aisles and stops at pick locations. The fundamental task of moving between picks is still performed – only now by a machine that moves slowly, is limited in its perception, and is prone to interruptions. Safety regulations cap the maximum speed of mobile robots to a level far slower than a trained picker walking with purpose. They wait at intersections, recalculate routes, stop for obstacles, and get bogged down in aisle congestion. While impressive from an engineering perspective, they are operationally underwhelming.
We are not reducing the number of travel movements – we are just automating them inefficiently.
Cost vs. Capability
Cost is another major concern. A single mobile picking robot costs between €30,000 and €80,000 depending on vendor and capabilities. This does not include:
- Annual software license fees (a very popular form of recurring revenue for startups that mistake themselves for Spotify)
- Fleet orchestration software
- Maintenance and technical support
- Battery management
- Physical safety measures
- WMS or ERP integration
In total cost of ownership (TCO) terms, a fleet of mobile robots can quickly become a sizeable investment with significant running cost. And still: your warehouse throughput will likely not exceed that of a lean, manual pick team operating in a warehouse with optimized SKU slotting under an wave-picking strategy.
Worse still, these robots cannot handle exceptions, errors, or ambiguous product situations. Their sensors are not perfect, and their ability to work with irregular goods, barcodes, or visual identification is limited. As a result, humans remain in the loop – to troubleshoot, assist, restock, or correct.
You also inherit a subtle shift in process responsibility: Instead of giving a task to a human who solves it end-to-end, you split the task across robot and human, creating more interfaces and handoffs. As always, every interface adds potential failure points and operational fragility.
So why, then, the continued interest?
If the above is true, why is there so much interest? I believe the reason is not operational – it’s psychological and political.
In my view, it is because mobile picking robots provide a lower-barrier narrative to automation. Mobile robots are tangible. They are “visible innovation.” They give the impression of progress without requiring the same fundamental process changes as a full GtP setup. They are easy to explain to management. A VP can walk into a warehouse with robots moving around and say: “We are automating.” It’s a convenient narrative.
From a CAPEX standpoint, mobile robots also present a more digestible investment: you can start with 5 or 10 units. You don’t need to redesign your warehouse entirely. You don’t need a mezzanine or structural modifications or a radically different fire safety concept. This makes them attractive in brownfield environments and for pilot projects. They feel like a “bridge” between manual operations and automation – and that’s an attractive pitch.
But this is a dangerous illusion. We risk falling into a technological dead-end where we incur the cost and complexity of automation without the performance benefits. And this, frankly, is worse than staying manual, because we lose both flexibility and efficiency. Moreover, the logic in favor of mobile picking robots only holds if the deployment is viewed in isolation – not in comparison to better alternatives.
Steelmanning the Case for Mobile Picking Robots
I must admit this isn’t going to be simple.
Before I try to steelman the case for mobile picking robots, I’d like to point out that there is a natural way out of the economic problem I outlined: if these things were really, really low-cost, I wouldn’t have a problem with the fact that they automate inefficient processes. If a resource is truly cheap and abundant, there’s little need to worry about process design and efficiency. That’s precisely why companies relocate production to low-cost countries where worker productivity may be lower, but wages are so low that it doesn’t matter. We’ll eventually get to that point with mobile picking robots as well, but we’re not quite there yet.
Okay, now back to my attempt at steelmanning.
If you normally operate only one shift but run mobile picking robots 24/7 to pick orders that are already in the system, the continuity can compensate for their lack of efficiency. This requires a significant time offset between when orders are received and when they need to be picked, which is a situation that is quite common for online retailers operating a single shift and relatively low customer expectations in terms of delivery time. Furthermore, it requires that a sufficient number of orders can be fully completed by the robots. An order structure with very few orderlines per order greatly facilitates this.
Another viable scenario might be one in which the alternative automated solution is excessively expensive. Enter automated case picking system. Automated case picking systems, as engineered and built by Witron (most of the time), SSI Schäfer (sometimes), and some others (almost never) are very expensive, require tons of space, and a total makeover of just about everything. These systems start at €50M (just logistics hardware and software, excluding building and everything else) and can reach €250M or more. At the same time, manual case picking is no walk in the park: it is dull, physically demanding, often even highly unergonomic, and companies frequently face labor shortages.
This is where the Linde MH’s RoCaP[1] and Solwr’s Grab[2] come in. These are AGVs equipped with robotic arms for fully automated case picking in brownfield environments. Linde’s RoCap has been under development for many years, the last four of which have been part of a joint research and development project with THWS (which happens to be the university where I teach). Admittedly, I was very skeptical of RoCap when I first heard about it – for all the reasons outlined above, and a few more. But given the enormous cost of conventional fully automated case picking systems, it might have been worthwhile to explore the idea. Whether it actually is worthwhile, I don’t know, as I’m not familiar with the cost structure of that machine. I could ask, but it doesn’t matter for the general point I’m making here: the alternatives are a manual process that is known to be terrible, and an automated process that is known to be excessively costly.[3]
I should add that I’m very familiar with case picking in food retail warehouses—this was my daily business for several years. And that’s precisely why I’m skeptical that case picking can be a process reliably operated by a mobile robot at a useful pace. The process is messy, variable, and sensitive to disruptions. But fast-forward 15 years, factoring in the technological progress we’re likely to see and some process adaptations that retailers would need to make to accommodate such systems, then perhaps it will work.
That’s as much as I can do in terms of steelmanning.
Pivoting to New Applications
For a client project, I was looking for an affordable solution for the replenishment of flow racks in a zone picking system. I also knew that several automotive companies were looking for a similar solution for the replenishment of flow racks at assembly lines, with some of them even building their own prototypes. The market was huge, and there wasn’t a single viable solution on the market.
Sometime right before COVID, I had a Zoom call with the CTO of a mobile picking robot company that was well known in Germany (it was impossible to find a conference in Germany where their leadership team wouldn’t be among the speakers), but mostly unknown elsewhere. I told him something along the lines of: “You’ve got great technology, but you’re wasting your time with the wrong application. Why don’t you replenish flow racks with your system?” He responded that this was precisely what they were piloting. This was great news. Checking their website today, five years later, line feeding is the first application presented to me.
If you are good with AGV technology, and you are good with vision and picking technology, pushing standardized plastic boxes into flow channels should be a much simpler task to accomplish. And there is a real market for it. Basically, any company operating an assembly line and having their line supply in flow racks could benefit from such a system. And many have explicitly demanded such a solution for years.
Mobile picking robots are a great example of a solution in search of a problem, while a real market was left unattended for most of the time, and for most companies out there, still is.
Concluding Thoughts
Mobile picking robots are a fascinating and often well-engineered technology. But in many warehouses, they solve the wrong problem. Instead of eliminating non-productive movement, they automate it – slowly and at high cost. It is not a sign of engineering ingenuity to automate a process that shouldn’t exist in the first place, or to replace one costly asset operating inefficiently with another costly asset that operates in the same manner but is far less capable.
As with all warehouse technologies, the key question is not can it be done, but should it be done, and in what context. Blind automation leads to fragile, expensive, and low-performance systems. Any attempt at automation should start with a clear diagnosis of the underlying process problem.
[1] https://www.kiongroup.com/en/News-Stories/Stories/Innovation/Innovative-Robotic-Truck-Successfully-Tested-with-ROSSMANN.html?storyid=59328 (last access: 2025-09-30)
[2] https://solwr.com/products/grab (last access: 2025-09-30)
[3] There is also a semi-automated option that is both terrible and costly. I’ll skip that one here for the sake of simplicity.
