Robots Atlas>ROBOTS ATLAS
Robotics & Hardware

Audi Tests Mobile Humanoids on Automotive Parts Sequencing

Lady Robot17 November 2026 · 7 min read
Audi Tests Mobile Humanoids on Automotive Parts Sequencing

Audi is testing mobile humanoids from Agile Robots for parts sequencing at its Production Lab in Ingolstadt. The project was announced on 21 October at the GTC conference in Berlin, with Dassault Systèmes and NVIDIA also taking part. Audi is another European carmaker bringing a humanoid into internal logistics — and another one doing so at proof-of-concept stage, with no results published.

Key takeaways

  • The task: picking components from containers and arranging them in order for assembly
  • The robots combine dual-arm manipulation and human-height reach with a mobile base
  • Audi's material-supply operations involve around one million picks daily — that is the scale of the existing process, not robot throughput
  • A virtual twin built in DELMIA on the 3DEXPERIENCE platform uses NVIDIA Omniverse libraries
  • The announcement describes a proof of concept, without quantified results or a factory-wide rollout timetable

What does parts sequencing involve?

An assembly line needs specific components in a specific order, matching the car variant currently moving along it. The parts, however, arrive at the plant loose, in large bulk containers.

Between the two stands a person who takes items out of the container and arranges them on a trolley or in a cassette in assembly order. That operation is called sequencing, and in automotive logistics it is one of the most labour-intensive steps.

According to Audi's announcement, its material-supply operations involve around one million picks daily. That number describes today's process, not the throughput the robots have demonstrated.

But here is the problem: the parts in a container lie unsorted, in random orientations, and some of them are floppy and change shape when grasped.

And exactly that set of conditions — varied, flexible and unsorted components inside an existing logistics environment — is what the partners name as the project's target.

What exactly does the robot do?

The Agile Robots machines combine three traits that together make the task possible. Dual-arm manipulation lets one hand hold a container or part while the other works. Human-height reach covers racks and bins designed around a worker. A mobile base lets the robot drive to the next station instead of working inside a single cell.

That last point is the meaningful break from classic automation. A stationary industrial arm requires the material to come to it. A mobile humanoid fits into a shop-floor layout designed for people, without rebuilding it.

Audi contributes the sequencing requirements and its operational expertise. Dassault Systèmes and NVIDIA cover the virtual-testing layer.

Why does this need a virtual twin?

The partners are building a virtual twin in DELMIA on the 3DEXPERIENCE platform, using NVIDIA Omniverse libraries. The model is to support checks on feasibility, timing, ergonomics and safety before wider deployment.

Each of those four points maps to a different kind of risk. Feasibility asks whether the robot can reach every position inside a container at all. Timing decides whether it finishes before the next car reaches the line. Ergonomics concerns a station shared with people. Safety is the precondition for letting the machine work among employees.

Testing this on the shop floor means occupying a real production line. In simulation it can be checked in parallel and more cheaply, though the classic problem of transferring results from simulation to physical hardware — sim-to-real — remains.

Logistics
Bulk container of parts
Robot
Humanoid picks a component
Does the part match the car variant on the line?
YES
Placed in sequenceAllow
NO
Pick another partDeny
Robot
Drive to the next station
Material ready for assemblyAllow

This is a simplified view built on the task description in the announcement, not documentation of a working installation.

Why is the car industry starting with sequencing?

Audi's project is not an outlier. The closest parallel is BMW and Figure 03 at the Spartanburg plant, where the task is practically identical: taking unsorted components from large containers and placing them into trolleys bound for assembly. BMW also runs a project with Hexagon's AEON humanoid in Leipzig, covering component manufacturing and high-voltage battery assembly among other applications.

Boston Dynamics is pursuing the same use case with Atlas at the Robotics Metaplant Application Center on Hyundai Motor Group's manufacturing campus in Georgia. The centre opened in September with an initial focus on automotive parts sequencing. Hyundai's roadmap targets factory deployment from 2028, which makes today's work a training effort ahead of production use.

The convergence is not accidental. Sequencing is repetitive, but not repetitive enough to justify building a dedicated machine for it. It demands no assembly-grade precision, it happens beside the line rather than on it, so a robot error does not stop production. On top of that, existing stations are already designed around human reach, which removes the cost of rebuilding the shop floor.

It is also worth noting what is missing from this set. None of these projects puts a humanoid on the assembly line itself, where cycle time is rigid and a mistake halts a whole shift. Internal logistics is a buffer in which a delay can be made up, and that is today's actual testing ground.

What do these names mean?

Four terms, without which the announcement reads as a list of proper nouns:

  • Sequencing — arranging parts in assembly order for one specific car. The equivalent of order picking in a warehouse, except the order is dictated by the production line.
  • Proof of concept — a pilot checking whether the solution works at all in a given environment. It implies neither a purchase decision nor a rollout schedule.
  • Virtual twin — a digital copy of a station in which a task flow can be tested without occupying the shop floor. The equivalent of a staging environment before a production deploy.
  • DELMIA on 3DEXPERIENCE — Dassault Systèmes software for planning and simulating manufacturing processes, where this twin is built.

The distinction between the second and third bullets decides how to read the announcement: tests run in simulation first, and only then is it settled whether the robot moves onto the floor more widely.

Is it working yet?

No, this is a proof of concept. The 21 October announcement describes a pilot at the Ingolstadt Production Lab, with no performance results and no factory-wide rollout timetable. The virtual twin is only meant to support checks before wider use, which in itself signals that the scaling decision has not been made.

Nor is it stated how many robots take part in the trial, what share of tasks they complete autonomously, or how often they need a human. The million picks a day is a parameter of today's human process, not measured machine throughput.

Why it matters

European carmaking is moving from demos to pilots at real stations, and doing so in logistics rather than on the assembly line. The choice of task says more here than the robot's specification: manufacturers are looking for a place where a humanoid can be wrong without halting production. Until announcements include cycle time and the share of autonomy, they confirm a direction rather than prove a business case.

What's next

  • The next checkable step is output from the virtual twin — feasibility, timing, ergonomics and safety are to be verified before wider deployment
  • Boston Dynamics is training Atlas on the same task at the centre beside Hyundai Motor Group's campus, with factory deployment planned from 2028
  • There is no timetable for extending Audi's project beyond the Ingolstadt Production Lab

Sources

Share this article