Developments & practice
How Stretch unloads cartons from containers at DHL
From stacked cartons to the conveyor: how Stretch unloads, retrieves fallen boxes and how to interpret DHL’s performance figures.
At DHL, the robot Stretch handles a physically demanding inbound task: it removes individually stacked cartons from containers or trailers and places them on a conveyor. DHL reports unloading rates of up to 700 cartons per hour. The figure comes from an announcement dated 13 May 2025 and describes an achieved peak, not a guaranteed average for every site.
DHL says it introduced Stretch commercially in North America in 2023, later extending deployment to the United Kingdom and other European locations. In May 2025, DHL and Boston Dynamics signed a memorandum of understanding for more than 1,000 additional units. The press release documents that intention to expand, not the completed installation of the announced number.
The case is particularly useful for understanding a robot’s role in an existing goods flow. The input is a loaded container. At the output, cartons sit on conveying equipment, ready for further identification and distribution. Between those points, the system must identify, reach and pick boxes arranged in different ways.
From a stack of cartons to the conveyor
In the DHL case study published by Boston Dynamics, employees describe the process: Stretch identifies cartons using cameras, picks them with suction cups and places them individually on the conveyor. They then pass through a scanning tunnel into the downstream system. Initially, an operator drives the robot into the container and enters information about the opening using a tablet.
The difficulty lies in the changing arrangement. Removing one carton changes the stack. Other boxes may shift or fall. Their previously observed positions are no longer valid. Before picking the next carton, the robot must account for the current situation. Its movement therefore depends on the conditions actually encountered.
In the case study, DHL explains that Stretch can also detect and retrieve fallen cartons. That matters to the workflow: if a person had to enter the area after every such event and reposition the box, unloading would be repeatedly interrupted. The sources do not, however, provide an intervention rate across sites. The described function therefore does not establish complete independence from human assistance.
In a separate technical article about retrieving fallen cartons, Boston Dynamics explains how this behaviour was improved. Previously, Stretch sometimes had to reverse and capture a new camera image because the fallen box was outside its previous field of view. The newer approach also uses measurements from laser sensors mounted on the base to determine the carton’s position.
These sensors generate a spatial collection of measurement points. Known surfaces, such as container walls and the conveyor, are filtered out of the search. Remaining clusters provide clues to the fallen box’s location. The system can use them to plan a pickup movement. The manufacturer describes faster recovery, but does not publish a specific before-and-after time in seconds in this article.
This technical explanation describes a product development. It does not automatically demonstrate that this software version was already in use at every DHL site mentioned. It remains important to distinguish the documented customer workflow from product functions described later. Both sources explain the same type of problem, but do not necessarily refer to the same installation version.
Performance depends on the complete unloading operation
The statement “up to 700 cartons per hour” is easy to misread without its frame of reference. Arithmetically, 700 cartons in 60 minutes corresponds to an average of about 5.1 seconds per carton. That does not mean every individual movement takes exactly that long or an entire shift runs at that rate. Preparation, changeovers and interruptions affect the total number processed.
An illustrative example clarifies the distinction. A system unloads for 50 minutes at a rate of 600 cartons per hour, processing 500 cartons during that period. If the operation, including other activities, takes 75 minutes overall, output across the complete period is 400 cartons per hour. Both 600 and 400 are mathematically correct, but refer to different time intervals.
A comparison should therefore define when measurement starts and ends. Does it count only the time when cartons are actually moving? Or does it begin with preparation and end when the unloading position is available for the next container? That choice changes the result. These example figures do not describe a DHL measurement; they show why the denominator belongs alongside every performance claim.
The downstream conveyor is just as important. If cartons cannot move onward after being deposited, faster picking offers little benefit. In its announcement, DHL mentions a UK project combining conveyors and palletisers in an end-to-end solution. Unloading is treated as part of a wider workflow that carries cartons through to a further organised staging step.
For this DHL case, the task, handover and a reported performance result are therefore documented. Stretch performs repeated lifting and placement, while the existing logistics system processes the cartons afterwards. The fallen-box example also shows why handling deviations is central to the job. A robust comparison between sites would additionally require average throughput, interventions and operating hours over a defined period.
Sources checked: 12 November 2026. Deployment and expansion figures refer to DHL’s May 2025 announcement. Additional manufacturer articles explain the process; the time-measurement calculation is fictional.