The recent wave of humanoid-robot announcements can be overwhelming, especially when every vendor claims a breakthrough. As an integrator, facility manager, or technology scout, you need a practical filter to separate operational reality from marketing momentum. The source material we reviewed covers three distinct developments that, taken together, give you a useful checklist for evaluating any humanoid robot you might deploy in 2025 and beyond.
First, look for evidence of continuous autonomous operation. The most striking data point comes from Figure AI, which reported that three of its Helix-02 humanoid robots crossed more than 24 hours of continuous autonomous operation in a package-sorting test. The test was originally scheduled for eight hours, but the robots kept going, and the company extended it. During that extended run, the robots sorted more than 28,000 packages without human control and without a reported failure. That is a concrete, measurable claim — not a vague promise. When you evaluate a robot, ask for the equivalent metric: How many hours of unattended operation has the vendor actually demonstrated? How many units ran in parallel? What was the throughput per hour? If a vendor cannot give you a number like “28,000 packages over 24+ hours,” you are looking at a concept, not a product.
Second, look for how the robot handles its own energy. The same Figure AI announcement includes a notable design choice: the robot can recharge itself wirelessly by standing on an inductive charging pad. The company’s third-generation humanoid, Figure 03, introduced in October 2025, is built around this self-charging capability. The design allows two robots to charge while others work, and the robot can swap its own batteries one at a time, staying powered indefinitely. This is a critical operational detail. If a humanoid robot requires a human to plug it in or swap batteries manually, then it is not truly autonomous — it is a machine that needs a handler. The inductive pad approach, while simple, is a meaningful step toward unattended operation. When you compare robots, ask about the charging architecture. Does the robot dock itself? Does it manage its own battery swaps? Can it operate through a full shift without a human touching it?
Third, look for evidence of deployment at scale in a real production environment. The source material includes a report from CATL, the battery manufacturer, which claims to be the world’s first battery maker to deploy humanoid robots at scale in battery pack production. The robots, named “Xiaomo” (or “Moz” in another translation), perform complex tasks like battery connector insertion. Crucially, the report states that the robot autonomously handles uncertainties such as material position deviations and connection point variations, dynamically adjusting its posture in real time. This is not a lab demo; it is a production line. When you evaluate a humanoid robot, ask whether the vendor can point to a named customer, a specific factory, and a specific task that has been running for months. If the only evidence is a promotional video, treat it as marketing.
Fourth, look for safety certification and the ability to work without cages. Agility Robotics is preparing to launch its Digit v5 humanoid robot in December, with a stated emphasis on working safely and cooperatively alongside humans without traditional safety fencing. The company’s chief business officer, Daniel Diez, said that what holds back large-scale deployments of humanoid robots is their inability to work outside safety parameters or safety cages. That is a candid admission of the industry’s biggest bottleneck. When you evaluate a robot, ask about its safety architecture. Does it have force-limited joints? Does it have proximity sensing that slows or stops the robot near a human? Has it been certified for human-robot collaboration without fencing? If the vendor cannot answer these questions, you will end up building cages around the robot, which defeats the purpose of a humanoid form factor.
Fifth, look at the business model and financial backing. Agility Robotics is going public via a $2.5 billion merger, with $300 million in booked revenue. That is a signal that the company has actual customers paying for actual deployments, not just venture capital funding for prototypes. When you evaluate a vendor, ask about booked revenue, not just total funding. A robot company with $300 million in booked revenue is a different risk profile than one with $300 million in venture funding and no sales.
Finally, look at the AI system that drives the robot. Figure AI’s robots run on Helix-02, an in-house vision-language-action AI platform. The Helix-02 system is what enables the robots to sort packages, adapt to variations, and operate for 24+ hours. The AI is the differentiator, not the hardware. When you evaluate a robot, ask about the AI stack. Is it in-house or third-party? What training data was used? How does it handle edge cases? A robot with a generic AI system will struggle in your specific environment.
Practical steps
Now that you know what to look for, here is a practical sequence of steps to evaluate and potentially deploy a humanoid robot in your operation.
Step one: Define your task precisely. The source material shows that humanoid robots are being used for specific, repetitive tasks: sorting small packages, inserting battery connectors, and handling material position deviations. Do not buy a humanoid robot for general-purpose work. Pick one task, define its parameters (object size, weight, cycle time, error tolerance), and ask the vendor whether their robot has demonstrated that exact task. If the vendor says “our robot can do many things,” ask for the one thing it has done in production.
Step two: Run a continuous operation test. The Figure AI test is your template. Start with an eight-hour shift, but plan for 24 hours. Measure throughput, error rate, and the number of human interventions required. If the robot needs a human to intervene more than a few times per shift, it is not ready for your floor. The source material notes that Figure AI’s robots sorted more than 28,000 packages in over 24 hours without a reported failure. Use that as a benchmark. If a vendor cannot match or exceed that level of reliability, keep looking.
Step three: Verify the charging and energy management system. Ask to see the robot charge itself. Does it navigate to a charging pad on its own? Does it manage battery swaps? The Figure 03 design allows two robots to charge while others work, and the robot can swap its own batteries one at a time. In your facility, you need to know how many charging pads you will need, how long a full charge lasts, and whether the robot can operate through a full shift without running low. If the vendor does not have a self-charging solution, you will need to factor in manual charging labor, which undermines the business case.
Step four: Check the safety certification and fencing requirements. The Agility Robotics Digit v5 is being positioned as a robot that works safely alongside humans without safety fencing. That is the goal. But you need to verify this for your specific environment. Ask for the safety certification documents, not just marketing claims. Ask about the robot’s behavior when a human enters its workspace. Does it slow down, stop, or reroute? What happens in a power failure? What is the emergency stop procedure? If the vendor cannot provide a clear safety case, you will need to budget for fencing, which adds cost and reduces the flexibility that makes humanoid robots attractive.
Step five: Inspect the AI system’s adaptability. The CATL deployment is notable because the robot autonomously handles uncertainties like material position deviations and connection point variations. That means the robot is not just following a fixed program; it is perceiving and adapting in real time. In your facility, you will have similar variations — boxes slightly out of place, connectors at slightly different angles, parts with minor defects. Ask the vendor how their AI system handles these variations. Ask for examples of edge cases the robot has encountered and how it responded. If the vendor cannot give you concrete examples, assume the robot will struggle with real-world variability.
Step six: Negotiate a pilot with a defined exit clause. Do not sign a long-term contract based on a demo. Negotiate a 90-day pilot with clear success metrics: throughput, error rate, uptime, and human interventions. Include an exit clause that allows you to terminate if the robot does not meet the metrics. The source material shows that Figure AI extended its test because the robots kept working — that is the kind of outcome you want, but you need to define it in advance. Also, ask about maintenance procedures. The source material notes that Figure AI says a robot can leave the work floor for maintenance if a software or hardware issue appears. That is a useful feature, but you need to know the expected maintenance frequency and the process for bringing a robot back online.
Step seven: Evaluate the vendor’s financial stability. Agility Robotics is going public via a $2.5 billion merger with $300 million in booked revenue. That is a sign of a company with real customers. When you evaluate a vendor, ask for their booked revenue, not just their funding. Ask about their customer list and reference sites. If a vendor cannot name a single production customer, you are taking on significant risk that the company may not exist in two years.
Step eight: Plan for the human impact. The source material raises the question directly: If robots can work through long shifts, what happens to the people who do this work today? You need a plan for your workforce. This is not just an ethical question; it is a practical one. You will need people to supervise the robots, handle exceptions, and maintain the systems. The source material notes that Figure AI’s robots can leave the work floor for maintenance if an issue appears — but someone has to diagnose and fix that issue. Plan for a transition where human workers are retrained to work alongside robots, not replaced overnight.
Step nine: Consider the total cost of ownership. The source material gives you some data points: Figure AI’s robots ran for 24+ hours, sorted 28,000 packages, and recharge wirelessly. CATL deployed robots at scale in battery production. Agility Robotics has $300 million in booked revenue. But none of these sources give you a price tag. You will need to ask the vendor for pricing, including the robot itself, the charging infrastructure, the AI software license, maintenance contracts, and training. Do not assume that a humanoid robot is cheaper than a traditional automation solution. The business case depends on your specific labor costs, shift patterns, and task complexity.
Step ten: Document everything. The Figure AI test was livestreamed, which gave people a front-row seat to real-time operation. That transparency is valuable. When you run your pilot, document everything: uptime, throughput, error rates, interventions, and maintenance events. Use that data to make your go/no-go decision. Do not rely on vendor claims; rely on your own measurements.
Common mistakes to avoid
The source material highlights several pitfalls that you can avoid with careful planning.
Mistake one: Believing that a 24-hour test means the robot is production-ready. The Figure AI test is impressive, but it was a test, not a production deployment. The robots sorted packages in a controlled environment. Your facility will have different conditions: different lighting, different floor surfaces, different package sizes, different noise levels. Do not extrapolate from a single test. Run your own pilot.
Mistake two: Ignoring the charging infrastructure. A robot that can recharge itself by standing on an inductive pad is a significant advance, but you still need to install the pads, manage the charging schedule, and handle the logistics of multiple robots sharing a charging area. The Figure 03 design allows two robots to charge while others work, but that implies you need at least three robots for continuous operation. Factor in the space and cost of charging infrastructure.
Mistake three: Assuming that “no safety fencing” means “no safety considerations.” Agility Robotics is positioning Digit v5 to work safely alongside humans without fencing, but that does not mean you can ignore safety. You still need to conduct a risk assessment, train your workers, and establish safety protocols. The robot may not need a cage, but your workers still need to know how to interact with it safely.
Mistake four: Overlooking the AI system’s limitations. The CATL deployment is notable because the robot handles uncertainties like material position deviations. But that capability is task-specific. The same robot may not handle a different task with different uncertainties. Do not assume that a robot that can insert battery connectors can also sort packages. The AI system must be trained for each task.
Mistake five: Underestimating the maintenance burden. The source material notes that Figure AI’s robots can leave the work floor for maintenance if a software or hardware issue appears. That is a useful feature, but it means you need a maintenance plan. Who will diagnose the issue? Who will fix it? What is the mean time to repair? The source material does not disclose these details, and you should not assume they are favorable. Ask the vendor for maintenance data.
Mistake six: Focusing only on the robot and ignoring the system. A humanoid robot is not a standalone device; it is part of a system that includes the AI software, the charging infrastructure, the safety systems, and the maintenance processes. The source material shows that Figure AI’s robots run on Helix-02, an in-house AI system. That means the AI is tightly integrated with the hardware. If you buy a robot with a third-party AI system, you may face integration challenges. Evaluate the entire system, not just the robot.
Mistake seven: Ignoring the financial risk. Agility Robotics has $300 million in booked revenue, which is a strong signal. But not every humanoid robot vendor has that level of commercial traction. The source material does not disclose the financial status of Figure AI or CATL’s robot program. Before you commit to a vendor, do your own due diligence. Ask for financial statements, customer references, and a roadmap for future development.
Mistake eight: Forgetting the human workforce. The source material raises the question of what happens to people who do repetitive tasks when robots can work nonstop. This is not just a philosophical question; it is a practical one. If you deploy humanoid robots, you will need to manage the transition for your workforce. The source material does not provide a solution, but you need to have a plan. Retrain workers for supervisory roles, maintenance, and exception handling. Do not let the deployment create a morale crisis.
Mistake nine: Assuming that “self-powering” means “zero energy cost.” The Figure 03 robot recharges wirelessly by standing on an inductive pad. That still requires electricity, and the pads need to be installed and maintained. The source material does not disclose the energy consumption or the cost of the charging infrastructure. Ask the vendor for these numbers before you commit.
Mistake ten: Rushing to deployment without a pilot. The source material shows that Figure AI extended its test because the robots kept working — but that was a test, not a production commitment. You should run your own pilot with your own metrics. The source material does not provide a template for a pilot, but you can create one based on the data points: 24+ hours of continuous operation, 28,000 packages sorted, no reported failures. Use those as your baseline, but set your own targets based on your specific task and environment.
Mistake eleven: Ignoring the regulatory landscape. The source material does not discuss regulations, but you should be aware that humanoid robots in the workplace may be subject to safety regulations, labor laws, and data privacy rules. The source material does not disclose any regulatory approvals for Figure AI, CATL, or Agility Robotics. Before you deploy, consult with legal counsel and regulatory experts to understand the requirements in your jurisdiction.
Mistake twelve: Assuming that a humanoid robot is the right solution for every task. The source material shows humanoid robots being used for specific tasks: sorting packages, inserting battery connectors, and working alongside humans. But a humanoid robot is not always the best solution. A traditional automation system may be cheaper, faster, and more reliable for a fixed task. The advantage of a humanoid robot is flexibility — the ability to adapt to different tasks and environments. If your task is highly repetitive and unchanging, a traditional system may be a better fit. Evaluate the alternatives before you commit.
Sources
http://electronics360.globalspec.com/article/22598/world-s-first-self-powering-humanoid-robot-works-nonstop
Published by Vigla Media OÜ (Estonia).