Article · AI translation
AI Takes Over Scheduling and Operations, and the Warehouse Gets Smarter as It Works
Embodied intelligence is beginning to take on more complex production settings.
While many companies are still exploring tasks such as picking and household chores, one player has now set its sights on a tough challenge in the industrialization of embodied intelligence: smart warehousing.
Its ambition is substantial: to have AI take over the entire warehouse, enable autonomous operation 24/7, and keep learning and evolving through real business operations.
That player is a familiar name: Geek+ (极智嘉), a leading global intelligent robotics company.
Recently, Geek+ officially unveiled its “24/7 Autonomous Smart Warehouse Plan,” aimed squarely at a self-evolving smart warehouse for the era of physical AI.
From getting embodied AI to do real work to getting an entire warehouse to run itself, physical AI is now confronting a deeper challenge in industrialization.
As AI moves deeper into industrial settings, an awkward fact has long remained hidden—
The more automated physical AI becomes, the more its reliance on people concentrates in the hardest parts of the work.
Over the past decade or more, AMRs have automated many standardized tasks, including transport and storage; goods can even find their own way to people.
But when it comes to complex picking, with thousands upon thousands of SKUs and packaging and placement that change frequently, robots still tend to struggle.
This also exposes a longstanding bottleneck in warehouse automation: gains in the efficiency of individual tasks can no longer easily support an intelligent leap for the entire warehouse.
It is at this juncture that Geek+ has offered its own industry assessment of where the next breakthrough in smart warehousing lies:
The key to the next generation of smart warehousing is to achieve autonomous 24/7 operation and continuous evolution, and to draw on commercial capabilities at global scale to make autonomous smart warehouses a new generation of productivity infrastructure.
The key to the next generation of smart warehousing is to achieve autonomous 24/7 operation and continuous evolution, and to draw on commercial capabilities at global scale to make autonomous smart warehouses a new generation of productivity infrastructure.
Behind this assessment lies an increasingly practical economic and technical calculation for the warehousing industry.
For customers, every additional hour a warehouse runs autonomously means another hour of capacity free from the constraints of human work shifts.
Especially at night and during order peaks, sustained and reliable fulfillment directly affects how quickly the supply chain can respond.
More importantly, it allows costly automation equipment to run at full capacity, turning the pressures of night-shift staffing and on-site management into tangible operational returns.
But that means robots must be able to handle anomalies they have never encountered, while the warehouse-wide system must be able to make decisions, recover from failures, and accumulate experience as hundreds of devices operate simultaneously.
After all, only when the entire warehouse can respond autonomously to uncertainty can technical autonomy truly become sustained, measurable productivity in commercial terms.
Based on this assessment, Geek+ has made warehouse-wide autonomy the strategic direction for the next stage of its “24/7 Autonomous Smart Warehouse Plan,” advancing it in 3 steps:
Connect the entire workflow: Let embodied intelligence take over complex operations and fill the gaps in warehousing tasks that have long relied on people.
Strengthen the intelligent feedback loop: Move AI from identifying problems to making autonomous decisions and dynamic adjustments, so the entire warehouse grows smarter as it works.
Deploy at scale: Use global delivery and localized operational capabilities to bring autonomous smart warehouses into more commercial settings.
These 3 directions correspond precisely to the 3 hurdles that physical AI must clear to be deployed at scale—
whether robots can perform complex tasks independently, whether the entire warehouse can make autonomous decisions, and whether this model can be replicated on a large scale.
In fact, this broad set of system-level capabilities has already taken tangible shape through the company’s years of technical accumulation and commercial practice.
For a warehouse to truly run autonomously, it must first solve a basic problem—
whether robots can connect the end-to-end chain of physical tasks, extending automation from isolated breakthroughs to the entire workflow.
How much work an individual robot can do determines the efficiency ceiling of a single task; whether the entire chain of operations can function autonomously as a closed loop determines how fully an unmanned warehouse can be put into practice.
Compared with embodied intelligence solutions focused on a single machine or specific task, Geek+, which has worked in embodied intelligence for more than 10 years, clearly has a deeper base of capabilities for real-world deployment.
Its self-developed Gravity unified embodied intelligence framework and Gravity 4D embodied world model target the challenge of autonomous robot operation in complex environments:
The former enables robots to understand tasks and plan actions, learning to think things through before acting; the latter enables them to understand geometric relationships in physical space and changes in force and contact.
Together, they give robots greater ability to respond autonomously to complex tasks and unexpected changes:
Generalization across tasks: When faced with unfamiliar items of different materials and shapes, robots can determine how to grasp them without training on each one individually.
Dynamic resistance to disruptions: If an item suddenly moves or becomes obscured, robots can adjust their actions in time and continue the task.
Long-horizon task execution: For complex tasks with multiple steps, robots can break them down, plan, and complete them step by step on their own.
Natural-language correction: If an action goes off course, the strategy can be adjusted through language instructions without reprogramming.
In addition, capabilities such as multimodal perception, long-horizon task reasoning, 4D physical modeling, and action planning are now bringing robots into real warehouse operations.
For example, its general-purpose humanoid robot Gino 1 can pick goods and move boxes, and can also coordinate with mobile robots to complete continuous picking tasks:
Meanwhile, Geek+’s unmanned picking workstation is also beginning to take over the picking and sorting stage, which previously depended heavily on people.
In a pilot at Schneider Electric’s Shanghai logistics center, the solution achieved picking efficiency 2 times that of human workers, with accuracy exceeding 99.99%.
Beyond these embodied intelligence products, its full range of AMR solutions also covers many standardized tasks involving different storage formats, including bins, pallets, and shelves.
The end-to-end technical capabilities built up over the years have also created favorable conditions for Geek+ to take the next step toward autonomous operation of the entire warehouse.
Getting robots to finish the work solves only the first challenge.
The bigger test is whether the entire warehouse can manage itself when hundreds of robots start working at the same time.
In a media interview, Tan Wenzhe, vice president of Geek+’s systems platform department, also discussed this problem that has long troubled the industry—
Traditional warehousing systems often each manage their own area: the scheduling system optimizes robot routes, the business system optimizes order execution, and the data platform handles statistics and reports.
Each system appears to have its own optimization goal, but in practice it is difficult to make unified decisions from the perspective of the entire warehouse!!
What Geek+ wants to do this time is give these previously separate systems a “warehouse brain” capable of coordinating the whole operation.
Specifically, this means bringing orders, inventory, robots, robotic arms, workstations, and more into a unified system for tracking operational status—
creating a complete feedback loop from physical operations and data sensing to intelligent decision-making, with real-time, warehouse-wide optimization.
At the foundation of the warehouse brain are 3 systems—RMS, WES, and IOP—responsible for robot scheduling, business execution, and operational analysis, respectively.
IOP identifies problems, WES formulates and orchestrates strategies, and RMS schedules robots to carry them out; the 3 work together.
What really makes this system smarter is the warehouse-wide intelligence Geek+ has built on top of it, using AI Agents to continuously strengthen 2 capabilities:
Collaborative intelligence: Schedule robots, equipment, and business processes together, flexibly adjusting tasks and resource allocation as orders change in real time.
Operational intelligence: Have AI monitor the entire warehouse in real time, diagnose anomalies autonomously, and optimize strategies, gradually reducing the need for human staff on duty.
Together, these mean AI must both manage the work at hand and keep figuring out how to do it better.
More crucially, the operational results of each adjustment can be fed back into the system, forming a complete “insight—diagnosis—decision—execution—learning” loop.
In this way, the warehouse begins to move beyond following instructions to finding problems and adjusting strategies on its own, growing smarter through long-term operation.
It is worth mentioning that Geek+ has already developed 3 types of AI Agent around this process—
Monitoring Agents identify anomalies, Data Analysis Agents investigate causes and propose solutions, and Deployment Configuration Agents adjust parameters and put strategies into practice.
These Agents also incorporate the “on-site operational experience” and “efficiency analysis frameworks” Geek+ has accumulated over the years.
Problems that experienced warehouse managers once had to investigate repeatedly now have a chance to become knowledge AI can access, analyze, and reuse, with optimization at different levels proceeding at different speeds.
In terms of efficiency, scheduling adjustments can even be made within a few minutes, while changes involving the overall operational strategy are continuously optimized using data accumulated each day.
The longer the warehouse runs, the more experience it accumulates, giving the system a chance to make better decisions—a true closed loop.
Competition to industrialize smart warehousing ultimately comes back to real-world settings.
What customers really care about is whether the system can run reliably over the long term, fit into their existing business, and deliver returns on their investment.
Geek+’s more than 10 years of commercial practice have already produced quite a few cases to examine.
In the global market, Geek+ has ranked first in AMR market share for 7 consecutive years. It operates in more than 40 countries and regions, serves over 1,000 customers, and has deployed more than 81,000 robots in total.
At this year’s CeMAT exhibition, Geek+ connected its full range of AMRs with AI scheduling across the entire warehouse, putting on a live demonstration of coordinated warehouse-wide operations.
In a media interview, Dr. Tan Wenzhe also gave several concrete metrics for assessing a warehouse’s capacity for “autonomous evolution.”
They include the ability to handle anomalies autonomously, throughput stability under complex operating conditions, the duration of continuous operation, and the efficiency of autonomous recovery after failures.
These metrics also point to the deeper significance of the “24/7 Autonomous Smart Warehouse Plan” for the industry:
gradually turning the tacit experience that veteran warehouse managers have accumulated over many years into intelligence that is “always present” in the warehouse.
For companies, few things are more frustrating than spending heavily to build an automated warehouse that still cannot function without people at critical moments.
A warehouse that can truly operate autonomously would mean that every bit of investment in automation can be turned into sustained, stable productivity.
Looking ahead, Geek+ also plans to extend its robot, AI, and system capabilities, validated in real-world settings, into a broader range of physical operations.
Along this path from technology to real productivity, the value of long-term accumulation and large-scale practice has become increasingly clear.