Contrary to some predictions, humanoids didn’t take over warehouses in 2025. But the robots sure took over the headlines.
DHL, Tesla, and Amazon made waves with plans to deploy humanoids in their facilities. NVIDIA CEO Jensen Huang took the CES stage flanked by a dozen bipedal automatons, demonstrating how robots can navigate complex warehouse environments using physical AI.
Most of these deployments are still in the pilot stage, and it remains to be seen how much value they’ll deliver. There’s no question humanoids are versatile. Unlike a robotic arm or an autonomous mobile robot (AMR), humanoids can move and manipulate the way a human would, and perform a range of different tasks the way humans do.
Nevertheless, I don’t expect such systems to replace existing human or robotic agents. Instead, they’ll find a more niche role: exception handling.
Large, automated warehouses already have specialized robotic systems for high-volume, repeatable work. Sorters, shuttles, AMRs, and conveyors move products at speeds a humanoid can’t match. The weak spot isn’t throughput. It’s recovery from disruptions.
A tote tips over at a merge. A misread barcode stops the line. A strap dangles in a sorter lane and triggers a fault. A human agent might have to walk the length of a football field just to locate the fix, leaving the zone idle and wasting precious uptime.
These occasional mistakes don’t justify custom automation. But they might justify keeping a dexterous robotic agent on hand. A humanoid can navigate to zones too dangerous for a human, use basic tools to reset the hardware quickly, and keep the flow moving.
What the case against humanoids misses
Skeptics see humanoids as “general-purpose” robots that don’t belong in warehouses at all. They point to cost, programming complexity, and vendor integration challenges.
After all, they argue, task-specific robots like conveyors, sorters, shuttles, and AMRs have proven economical and reliable time and time again. When you’re operating on tight service-level agreements (SLAs), why roll the dice on general-purpose robots when you already have a fleet executing high-volume tasks at scale?
It’s true that humanoids can’t keep the same pace as specialized warehouse automation. But a humanoid doesn’t need to pick 400 units per hour (UPH). That’s what automated storage and retrieval systems (ASRS) are for.
You bring in the humanoid when you need to clear a jam or reset a fault in minutes, so the ASRS doing 400 UPH doesn’t sit idle. The return on investment (ROI) of humanoids comes from uptime across the system. Leave the high-volume work to task-specific robots, and give the exception handling to humanoids.
Physical AI breakthroughs make humanoid exception handling possible
The potential for humanoid exception handling is grounded in two major developments in physical AI. First, rapid progress in neural network training and environmental sensing has made it possible for humanoids to work safely in mixed environments.
Second, AI-driven warehouse orchestration software can coordinate thousands of fixed, mobile, and humanoid agents in real time, learning how they behave in real-world conditions to optimize their paths and task execution.
A decade ago, routing 1,000 robots from Point A to Point B required hours of compute for a single scenario. Today, trained models can generate safe, efficient paths in milliseconds after running millions of simulated episodes. That speed allows the system to detect a stalled tote, classify it, and route the nearest capable agent — including a humanoid — without halting other work.
The same warehouse orchestration layer that assigns AMRs to replenishment tasks can dispatch a humanoid to clear a fault, then resume normal traffic patterns as soon as the block is removed. No separate control system. No major downtime.
What humanoids should (and shouldn’t) do in automated warehouses
As we’re still piloting humanoids’ performance and scalability in commercial deployments, investing in specialized hardware and orchestration software that can optimize workflows and execution remains the safest way to boost warehouse productivity for now.
Task-specific machines like AMRs, conveyors, arms, and cobots will continue to handle the bulk of fulfillment. But when those systems fail, a humanoid may be able to intervene more quickly and safely than a human can. And AI-driven orchestration can direct all agents, humanoid or otherwise, to where they’re most useful.
We won’t be giving humanoids the key to our warehouses. We’ll just be calling them in when we need a hand.