A chatbot can explain how to move a heavy cart. A robot has to get underneath it, carry the weight and avoid the person walking across its path.
Amazon is working on a machine that connects those two worlds. Its next-generation Proteus is designed to receive instructions in everyday language and turn them into movement around a warehouse.
Unveiled on 4 June 2026, the system is still being piloted in Amazon’s laboratories, with European deployment planned for the first half of 2027. It is a development programme with a stated timetable, not a feature already installed throughout the company’s network. Amazon’s announcement explains that distinction.
How Amazon Proteus follows text instructions
The idea is straightforward: an employee specifies a task, and the robot works out how to carry it out. Amazon’s broader announcement explicitly describes conversational text-based prompts. So this is not a confirmed claim that every worker can simply shout across the warehouse and summon a robot. Source: Amazon’s detailed overview.
For example, imagine asking a machine to move a particular container to a particular workstation. That illustrative instruction sounds simple. Completing it requires identifying the right container, understanding the destination, choosing a route and fitting the job around other activity.
The promised improvement is that the person can express the goal without entering a specialist sequence of commands. That is a meaningful interface change, even if the machine doing the work looks more like a moving platform than a science-fiction android.
The robot does not need a human shape
The original Proteus transports heavy carts in dock areas. Amazon says it is deployed at 25 US fulfilment centres and handles carts that can weigh close to 400 kilograms. The proposed successor extends that role to more places where containers need moving, including transfers between workstations. Source: Amazon’s Proteus announcement.
A low platform can make more sense than a humanoid for this job. It needs stability, dependable navigation and the ability to carry a load. Two legs, a face and human-looking fingers would not automatically make it a better cart mover.
That is one reason everyday robotics may arrive less dramatically than the films suggest. The useful machines could disappear into the background, doing a narrow job repeatedly rather than trying to imitate an entire person.
Another robot is learning when to stop squeezing
Moving a cart is different from picking a product out of a crowded shelf. Amazon’s Vulcan system addresses that second problem using force feedback: sensors help it detect contact and adjust how hard it pushes or grips.
Amazon describes tools that make space inside storage compartments and a camera-and-suction arrangement for retrieving items. The company says Vulcan can handle roughly 75% of the types of items in its fulfilment centres, while referring unsuitable items to people. That is a company-reported capability, not evidence that a robot can handle any object in any home. Amazon’s Vulcan explanation describes its strengths and limits.
Vulcan is already being used in facilities including Spokane and Hamburg. One intended benefit is handling high and low storage positions that would otherwise involve more climbing or bending. The future expansion of Proteus and the existing use of Vulcan should not be confused: they are different systems at different stages.
What changes for people?
For customers, the potential benefit is less waiting while an order moves through the building. But a faster robot cannot remove every bottleneck between checkout and the doorstep.
For employees, the bigger questions concern physical workload, training, job design and how productivity gains are shared. Amazon frames its robotics programme as support for workers and says its broader European expansion includes plans for 25,000 additional fulfilment jobs. Those are company plans; they do not establish how every individual role will change. Source: Amazon’s investment overview.
Conversational commands also create a practical test: what happens when an instruction is ambiguous? A useful system must know when to ask for clarification, pause or hand the task back—not merely produce a confident interpretation.
The most interesting promise here is modest enough to be believable: people describe what needs doing, while specialised machines take on more of the physical effort. Whether that becomes a better workplace will depend on the deployment, not on how naturally the instructions are written.
For another specialised approach to robotics, read about Roborock Saros Rover and stair-climbing robot vacuums.
Leave a Reply