Humanoid Robots Doing Paid Work: Digit Proof

The clearest current answer is Agility Digit. Agility now has public commercial agreements with Toyota Motor Manufacturing Canada and Mercado Libre, says Digit has moved more than 100,000 totes at GXO's Flowery Branch facility, and describes Digit as commercially deployed with customers including GXO, Schaeffler, Amazon, Toyota, and Mercado Libre. That does not make Digit a home robot. It does make Digit one of the best public tests of whether humanoid robotics has moved beyond stage demos.

For home robot buyers, this matters because factory and warehouse deployments are the proving ground before household claims become believable. A robot that cannot survive repetitive paid work in a controlled facility is not ready to fold laundry, carry dishes, work around pets, and recover from everyday home chaos.

Which humanoid robots are already doing paid work?

There are several levels of evidence. A humanoid robot can have a press demo, a pilot, a commercial agreement, a paid deployment, or measured production work. Those should not be treated as the same thing.

The pattern is obvious: the best public evidence is not a cinematic video. It is a named customer, a paid or commercial relationship, a specific worksite, a task that happens repeatedly, and metrics that include time, volume, failures, or interventions.

That is why Digit is the useful case study. It has named customers, a repeatable logistics task, a fleet software layer, charging and safety updates, and enough public detail to ask serious buying questions.

What does Digit's deployment prove for home robot buyers?

Digit proves that a humanoid form factor can be useful in some human-designed workplaces without rebuilding the whole environment around a robot. Agility's official Toyota announcement says Toyota Motor Manufacturing Canada signed a commercial agreement after a successful pilot and plans to deploy Digit in its facilities for manufacturing, supply-chain, and logistics operations.

The Mercado Libre agreement is even more explicit about logistics. Agility says Digit will enter Mercado Libre's San Antonio, Texas facility, initially focusing on commerce-fulfillment tasks, with the companies exploring additional logistics use cases across Latin America. The same announcement says Digit has already moved more than 100,000 totes in live commerce operations.

Those are not home tasks. They are controlled industrial workflows. But they are exactly the kind of proof that should come before home marketing claims:

• A customer pays or signs a commercial agreement. That is stronger than a founder saying the robot is nearly ready.
• The robot repeats boring work. Totes, parts, and workcells expose uptime, calibration, gripper wear, networking, charging, and support issues.
• The deployment has a support stack. Agility positions Digit with Agility Arc, a cloud automation platform for fleet management, workflow monitoring, AMR coordination, and enterprise-system integration.
• Safety is part of the product, not an afterthought. Agility's 2025 Digit update added Category 1 stop behavior, safety PLC support, on-robot E-stop, wireless teach-pendant E-stop, and Functional Safety over EtherCAT.

The takeaway for home buyers is not "buy Digit for the house." ui44 lists Digit as an active humanoid with enterprise deployment only, no public price, and an RaaS/contact-sales model. The takeaway is that paid work gives buyers a better question set. If a future home humanoid company cannot explain deployment history, support, charging, uptime, intervention rates, and safety boundaries, it is asking for trust before it has earned it.

The numbers that separate proof from hype

The ui44 database keeps the comparison grounded. Digit is not cheap consumer electronics. It is a 175 cm, 65 kg humanoid with about 4 hours of battery life, roughly 2 hours of charging time, 3.4 mph maximum speed, LiDAR, RGB-D cameras, force sensing, and Agility Arc as its planning and fleet layer. Its listed capabilities include 16 kg box carrying, autonomous navigation, warehouse operations, stair navigation, and fleet coordination.

That spec sheet explains why warehouse work is the starting point. A 65 kg biped carrying boxes around people needs managed routes, task rules, safety systems, operators, maintenance, and charging plans. Homes need all of that too, but with less structure and less tolerance for mistakes.

Figure's BMW data shows what better proof looks like. Figure says Figure 02 ran 10-hour shifts Monday-Friday, loaded 90,000+ parts, accumulated 1,250+ hours of runtime, contributed to 30,000+ BMW X3 vehicles, and logged about 1.2 million steps. It also published the specific assembly-line KPIs: 84-second total cycle time, 37-second load time, greater than 99% target placement success per shift, and zero human interventions per shift as the goal.

That is the right direction for humanoid transparency. A buyer can inspect the claim, compare it with the robot's status, and ask what changed in the next generation. In ui44, Figure 02 is now discontinued, while Figure 03 is active with no public price, about 173 cm height, 61 kg weight, about 5 hours of battery life, and a 20 kg payload. The important point is not that Figure 03 is a home product today. It is that Figure can point to a real industrial learning loop behind the next robot.

Digit's public proof is similar in spirit but different in task. The GXO tote milestone is about long-running logistics throughput. Toyota and Mercado Libre are about named commercial expansion. Agility Arc is about fleet operations. Put together, they make a stronger case than a one-off humanoid dance, speech demo, or carefully edited kitchen clip.

A paid deployment still is not home readiness

A warehouse robot gets help from the environment. The bins are standardized. The workflow is repeated. Lighting, floor surfaces, object classes, paths, and handoff points are relatively controlled. A home is the opposite. It has cables, loose clothes, pets, stairs, narrow furniture gaps, family preferences, fragile objects, and social rules that are hard to encode.

That is why paid industrial work should be treated as a necessary but not sufficient step toward home robots. It can prove hardware durability, manipulation repeatability, support economics, and safety discipline. It cannot by itself prove that a robot can decide whether a glass belongs in the sink, a dog toy belongs in a basket, or a child's school paper should never be thrown away.

The gap shows up in pricing and support too. Digit has no consumer price in the ui44 database. Apptronik Apollo has no public price. Figure 03 has no public price. UBTECH Walker S2 has no public price. Sanctuary Phoenix has no public price. Even 1X NEO, the rare home-focused humanoid with a listed early-adopter price in ui44, is a $20,000 pre-order rather than a normal mass-market appliance.

For a home buyer, that means the honest question is not "which humanoid can do paid work?" It is "which company's paid-work evidence transfers to my home?" A warehouse tote workflow transfers some lessons: uptime, grasping, charging, fleet monitoring, and support. It does not transfer household judgment.

How to read the next commercial humanoid announcement

When a company announces a humanoid deployment, look for five details before you believe the headline.

1. Is there a named customer? "A Fortune 500 company" is weaker than Toyota, Mercado Libre, BMW, Mercedes-Benz, GXO, Schaeffler, or Amazon. Named customers put reputation behind the claim.

2. Is it a pilot, commercial agreement, paid deployment, or scaled rollout? A pilot can be valuable, but it is not the same as recurring paid work. A commercial agreement is stronger, and measured customer throughput is stronger again.

3. What is the repeated task? "General purpose" means little without a task. Digit's useful public examples are totes, logistics handoffs, AMR integration, and manufacturing support. Figure's BMW proof is sheet-metal loading. Apollo's Mercedes-Benz use case is assembly-kit and tote delivery. Those are specific enough to evaluate.

4. What are the boring metrics? Hours, cycles, parts moved, totes moved, interventions, charging, failed grasps, emergency stops, and maintenance windows matter more than a dramatic clip. If a company will not share any denominator, the evidence is weak.

5. What happens when the robot fails? Home buyers should care deeply about recovery policy. Does the robot stop safely, retry, call an operator, hand off to a human, or keep pushing? Industrial customers demand this because downtime costs money. Homes need it because mistakes can become safety problems.

This is where Digit's industrial focus is instructive. Agility talks about cooperative safety, fleet monitoring, remote support, integration with AMRs, chargers, workcell emergency-stop systems, and enterprise software. Those are not glamorous details, but they are exactly the details a future home robot needs in a consumer-friendly form.

Why charging and uptime may matter more than walking

Humanoid marketing loves walking. Buyers should pay equal attention to charging. A robot that can walk for four hours and then needs supervised recovery is not a worker; it is a demo schedule.

Digit's current ui44 entry lists about 4 hours of battery life and about 2 hours of charging. Agility's 2025 update says Digit gained more efficient battery capabilities up to four hours plus autonomous docking onto a charging station. That matters because a paid deployment needs shifts, not clips.

UBTECH Walker S2 pushes the same issue from another direction. Its official product page centers the claim that Walker S2 can swap its own battery in about 3 minutes, uses a dual-battery design, and targets 24/7 continuous operation. ui44 tracks Walker S2 as an active industrial humanoid with 15 kg payload, RGB binocular stereo vision, BrainNet 2.0 plus Co-Agent task planning, and no public price.

The home version of this question is simple: can the robot keep itself useful without becoming another device the owner has to babysit? Autonomous docking, battery swapping, diagnostics, remote support, and predictable service plans may sound less exciting than walking up stairs. They are probably more important for real ownership.

The practical buyer checklist

If you are watching the humanoid robot market from the home side, use paid-work evidence as a filter, not as a purchase signal.

A serious future home humanoid should be able to answer these questions:

• Which paid deployments taught this robot to survive real use?
• How many hours, cycles, objects, rooms, or tasks are behind the autonomy claim?
• What is the public intervention rate, and how is an intervention defined?
• Can the robot charge, dock, update, diagnose, and recover without constant owner supervision?
• What is the worst task it is allowed to attempt around people?
• Who services it when a joint, sensor, hand, battery, or camera fails?
• Does the company publish a consumer price, warranty, privacy policy, and safety boundary, or only enterprise contact forms?

Digit gives unusually strong answers for the industrial version of those questions. Figure 02 gives strong measured-runtime evidence from BMW. Apptronik, UBTECH, Sanctuary, and 1X each show pieces of the path. None of that turns a humanoid into a normal home appliance yet.

The buying lesson is clear: treat paid work as the minimum credibility layer. The best humanoid robot demos are no longer the flashiest ones. They are the ones that show boring work, named customers, measured throughput, recovery, charging, and support. That is the evidence trail home robots will need before "help around the house" becomes more than a promise.