ROKAE Names Five Helios Robots at Schaeffler's Humanoid Factory

How the Taicang Factory Connects Robots and Components

Schaeffler's October 20 factory announcement describes an assembly line, training and data acquisition, digital twins and a data center at one site. Its stated purpose includes manufacturing linear and rotary actuators, the assemblies that turn electrical power into controlled motion inside robot joints.

The customer describes six humanoids working across five stations, covering preparation, sorting, joining, inspection, labeling and packaging. ROKAE's later release identifies five Helios machines equipped with its AR-series arms. The sources describe different scopes; they do not identify every robot on the line or map each station to a particular supplier. There is no basis to assume that six and five are interchangeable fleet totals.

The training role raises a practical measurement question. A robot can improve at a task while the manufacturing process still relies on carefully prepared inputs or frequent human resets. Useful reporting would distinguish training runs from accepted production work and say which conditions changed during evaluation. Without that separation, a longer demonstration may look like an increase in useful factory capacity even when it measures a different activity.

Under the Hood: Force Sensing Before Fleet Intelligence

Force control adjusts a robot's motion using measurements of contact rather than relying only on a programmed position. When parts meet during assembly, a small alignment error can create resistance. A controller can use that feedback to adjust the movement or limit the applied load. The difficult part is making those responses dependable across the full task, including unexpected contact and recovery.

ROKAE's AR product specifications list three arm sizes. Each has seven degrees of freedom, meaning seven independently controlled joints. The extra joint gives an arm more freedom to choose a posture while reaching the same tool position, although the usable workspace still depends on the body, tooling, cables and surrounding equipment.

These are supplier specifications, not measurements from Taicang. The public materials do not identify which arm variant each installed robot carries. A payload rating also cannot be treated as a complete assembly capability: tool mass, reach, acceleration and contact forces all affect what a particular installation can do.

The AR technical sheet separates joint torque sensing from an end-mounted force sensor. It lists joint sensing as standard and describes force-control accuracy figures as reference values. Torque is the turning load at a joint; measuring it helps the controller estimate interaction along the arm. An end sensor instead measures loads near the tool. Neither specification is a published failure rate for inserting, fastening or inspecting a real component.

Helios adds mobility and a second arm. Its current product table lists Helios II variants with 5 kg or 10 kg payload per arm and an omnidirectional wheeled base. Those catalog options should not be assigned to the Schaeffler installation without a configuration disclosure. Wheels can suit level factory floors; this deployment supplies no evidence about bipedal walking.

ROKAE and Schaeffler: A Supplier Relationship in Both Directions

The installation follows an April 15 partnership announcement. ROKAE said the companies would co-develop integrated humanoid joint modules, combining its control and integration work with Schaeffler's torque sensors and harmonic drives. The agreement also envisaged Schaeffler supplying components and ROKAE participating in manufacturing automation.

That gives the relationship two possible business directions: supplying the parts inside robots and supplying robots to help manufacture parts. The Taicang disclosure provides a named example of the latter. It does not reveal the contract value, ownership of the machines, payment structure, service commitments or whether the installed equipment uses the jointly developed module.

The arm milestone has a similarly precise boundary. ROKAE's September 29 exchange filing defines cumulative roll-off as output since that product line entered production. It is not an annual production rate, an installed humanoid count or a measure of active customer utilization. October's delivery wording does not supply an audited reconciliation between factory output, shipment and customer acceptance.

What This Means for Factory Buyers

A component supplier has a different scaling problem from a complete-robot vendor. An arm may be purchased for a fixed workstation, a mobile manipulator, data collection or a humanoid body. The same component family can therefore reach customers whose final applications and levels of autonomy vary considerably. Component volumes are useful evidence of manufacturing activity, but cannot establish demand for one particular robot form.

For a factory operator, the decisive unit is useful output per operating hour. A credible assessment would count accepted parts, rejected work, unplanned stops, recovery time and the people needed to keep the process running. It would also include commissioning, fixtures, software integration, maintenance and spare parts. Hardware throughput alone leaves those costs unresolved.

The most useful next disclosure would connect a defined task to a measured result over a stated period. For example, insertion success measured only on prepared parts would answer a narrower question than full-shift assembly with mixed incoming tolerances. Both can be useful, provided the conditions and denominator are clear.

Recovery deserves its own measure. If an arm encounters a misaligned part, the outcome might be a successful retry, a rejected component or a call for an operator. Counting only completed cycles would conceal the time spent resolving those exceptions. A deployment report should therefore explain the recovery policy and count interventions over the same period as its production total.

The two-arm configuration creates another evaluation boundary. Coordinated handling can be useful when one arm holds a part while the other performs work, but it also introduces possible collisions between tools, limbs and fixtures. The relevant evidence is the complete configured task, including stopping behavior and restart procedures. A component's sensing specification cannot answer every question about the assembled workstation.

What's Coming Next

The October 26 release is a specific supplier update, rather than proof that flexible factory work has been solved. Its value will grow if the companies disclose installed configurations, task allocation and operating results that can be compared over time.

Expansion beyond this line would be another meaningful test. Repeating a task in a second plant can reveal how much engineering is reusable and how much depends on local fixtures, layouts and production rules. Buyers should watch for repeat installations with comparable performance, together with evidence that maintenance and recovery remain manageable as the fleet grows.

Frequently Asked Questions

The Next Measure Is Accepted Work

A named installation makes the humanoid supply chain more tangible. The stronger evidence will be a record of usable components assembled consistently, with the supervision and costs included. That is the bridge between manufacturing robot hardware and manufacturing economically with it.

Hero photograph: Schaeffler KG's headquarters in Herzogenaurach, Germany, archival image uploaded in 2006. It does not show Taicang or a Helios robot. Schaeffler KG / Wikimedia Commons; public-domain release confirmed through Wikimedia VRT ticket 2007101610005246. Display resizing and cropping may apply.