Geek+ unveils 24/7 autonomous warehouse plan with AI running operations

Geek+, the Chinese warehouse robotics company, has unveiled a plan it calls the "24/7 Autonomous Intelligent Warehouse," positioning full-warehouse autonomy as its next strategic direction. The goal is not to make a single robot more productive but to let an entire warehouse run continuously without staffed shifts and keep learning from live operations.

The plan is laid out in three phases. The first is connecting the full workflow, so embodied AI can take over complex tasks that still depend on people. The second is tightening the intelligence loop, moving AI from spotting problems to making decisions and adjusting on its own. The third is scaling the model through Geek+'s global delivery and local operations network. The company frames these as the three hurdles physical AI must clear: can a robot complete complex work independently, can a whole warehouse decide for itself, and can the model be replicated at scale.

The technical foundation is Geek+'s in-house Gravity unified embodied intelligence framework and Gravity 4D embodied world model. The first handles task understanding and motion planning; the second models geometry, force and contact in physical space. Together, the company says, they give robots four capabilities: generalizing across unfamiliar objects without training each one, adjusting when items move or are occluded, decomposing and executing long multi-step tasks, and correcting a motion through natural-language instructions instead of reprogramming.

On the hardware side, Geek+'s general-purpose humanoid Gino 1 can pick goods, move cartons and coordinate with mobile robots on continuous picking runs, while an unmanned picking workstation takes over the sort-and-batch step. In a pilot at Schneider Electric's Shanghai logistics center, Geek+ says the system reached twice the picking efficiency of manual work with accuracy above 99.99%.

Getting robots to finish tasks is only the first test. The harder problem is whether a warehouse can manage itself when hundreds of machines are running at once. Tan Wenzhe, vice president of Geek+'s system platform division, described the industry's familiar fragmentation in a media interview: scheduling systems optimize robot routes, business systems optimize order fulfillment and data platforms produce reports, each with its own objective and none able to decide for the warehouse as a whole.

Geek+'s answer is a warehouse brain that puts orders, inventory, robots, robotic arms and workstations into one operating-state system. Three platforms sit underneath: RMS for robot scheduling, WES for business execution and IOP for operational analysis. On top of that, a monitoring agent flags anomalies, a data-analysis agent traces causes and proposes fixes, and a deployment agent applies parameter and strategy changes. Scheduling adjustments can land in minutes, while broader operating strategies are refined daily, feeding results back into an "insight, diagnosis, decision, execution, learning" loop.

The commercial base behind the pitch is substantial: Geek+ says it has led the AMR market for seven consecutive years, serves more than 1,000 customers across over 40 countries and regions, and has deployed more than 81,000 robots. Tan's yardsticks for "autonomous evolution" are deliberately practical: handling exceptions unaided, throughput stability under difficult conditions, continuous runtime and recovery speed after a fault.

What to watch next is whether round-the-clock autonomy moves from showcase projects to repeatable deployments. For customers, every extra hour of unattended operation is capacity free of shift limits and a way to get more out of expensive automation, but turning night-shift and site-management costs into measurable returns depends on how reliably the system handles exceptions and recovers from failure. The competitive question in physical AI is shifting from how much one robot can move to whether an entire warehouse can run itself.