Figure AI put its humanoid robots through a grueling test last week: sort 88,000 packages in 72 hours, all while streaming the operation live. The company’s bots pulled it off, but the demonstration also laid bare a challenge that could slow real-world adoption — accuracy isn’t perfect yet.
The 72-hour sorting marathon
The livestream showed multiple humanoid robots working around the clock at a logistics facility. Over three days they handled 88,000 parcels, a pace that would translate to roughly 29,000 packages per day. Figure AI designed the test to showcase what its robots can do in a warehouse setting, where labor shortages and rising e-commerce volumes keep pressure on operators to automate.
Sorting packages is a repetitive task, but it’s not simple. Bots have to identify, pick, and place items of varying shapes and sizes without damaging them. Figure AI’s humanoids managed the volume, but the company acknowledged that accuracy issues remain. The robots occasionally mis-sorted or dropped items, errors that add up in a commercial environment where even a 1% mistake rate means hundreds of misdirected packages per shift.
Why accuracy matters for logistics robots
Logistics companies operate on thin margins. A sorting error can mean a customer gets the wrong product, a return gets mishandled, or a shipment is delayed. For a robot to be commercially viable, it has to match or beat human accuracy — and human sorters still hit 99.9% or better in many facilities. Figure AI hasn’t released detailed error rates from the livestream, but the fact that the company itself flagged accuracy as a hurdle suggests the numbers aren’t yet where they need to be.
That’s a common pain point for humanoid robots. Unlike fixed industrial arms, humanoids are general-purpose machines meant to move through a space and handle varied tasks. The trade-off is complexity: more joints, more sensors, more software, more points of failure. Figure AI’s test proved its robots can sustain a high workload, but consistency over thousands of repetitions is a different metric.
The path to commercial deployment
Figure AI has positioned itself as a leader in humanoid robotics, with a focus on logistics and warehouse automation. The 72-hour livestream was a publicity push, but it also served as a real-world stress test. The company likely gathered data on failure modes, recovery times, and system reliability under continuous operation — all critical for any company thinking about buying these machines.
Potential customers will want to see error rates drop before they place orders. A robot that sorts 88,000 packages in 72 hours is impressive, but one that mis-sorts 500 of them is a liability. Figure AI hasn’t said when it plans to start commercial shipments or how much the robots will cost. The accuracy issues raised in the demonstration are the kind of detail that determines whether a pilot project turns into a fleet deployment — or stays a demo.
For now, the company has shown its robots can work long hours. The question is whether they can work them well enough.
