Material picking, positioning, insertion, and feeding.
At the recent China International Import Expo, an A3 robot from Tars demonstrated autonomous connector assembly at the Aptiv booth. This routine isn't just for show; the same model has already entered Aptiv's actual production lines.
Image source: Tars
That day, Aptiv signed cooperation agreements with Tars and Shanghai Electric. According to the plan, hundreds of embodied robots from both partners will gradually enter Aptiv's production scenarios. Tars revealed plans to deploy hundreds of A-series robots by the end of 2026, covering eight major production processes and 33 application scenarios.
Conversely, Aptiv's high-performance computing platforms and intelligent power management technologies will also be integrated into its partners' robot products.
Yang Xiaoming, President of Aptiv China and Asia Pacific, summed up the relationship during an interview at the expo: "We act as both their component supplier and their customer."
Beyond buying and selling, the collaboration involves a resource that isn't easy to put a price tag on: Aptiv's own production lines.
One Production Line, Two Deals
Auto parts suppliers selling products to the robotics market is hardly news anymore.
Wang Xianbin, a partner and vice president at the Gasgoo Research Institute, previously noted that the overlap between automotive and robotics supply chains is about 60% to 70%. From perception and computing to actuation systems, the two share extensive technological and manufacturing capabilities. With adaptive development, the parts technology, cost control, and mass production experience accumulated by the auto industry can migrate to the robotics sector.
Multinational Tier 1 suppliers like Schaeffler, Bosch, and Valeo have already expanded their businesses along this path.
Now, they are bringing their production lines to the negotiating table as well.
Take Aptiv as an example: the first batch of Taishi A-series robots has entered actual production lines for process adaptation and application verification. Both sides will continue to accumulate data around industrial scenarios to support the iterative upgrade of Taishi's general embodied large model, AWE.
Aptiv, meanwhile, is applying its own computing platforms and power management solutions to robot products, with mass production planned for later this year.
Image source: Tars
Robotics companies provide equipment to Aptiv, while Aptiv supplies core technologies to them. The production line simultaneously becomes a venue for robot verification and training.
This two-way cooperation holds another layer of value.
Yang mentioned in the interview that beyond simply substituting human labor, data generated by robot operations can be used to improve capacity planning, quality tracking, and equipment prediction—thereby helping factories boost efficiency and cut costs.
In other words, Aptiv cares not only whether robots can handle specific production tasks, but also whether their integration brings additional data and management capabilities to the existing manufacturing system.
Schaeffler has taken a different path.
On October 20, Schaeffler's Taicang Digital Intelligence Factory for embodied intelligent robots officially opened. This roughly 7,500-square-meter facility integrates functions including robot parts manufacturing, scenario training, data collection, and digital twins. Six humanoid robots participate across five workstations, handling tasks such as sorting, press-fitting, screw assembly, and inspection.
Here, robots participate in manufacturing components needed for the next generation of robots, and their actual operations become part of the training and engineering verification process.
Schaeffler itself is also supplying core products like joint actuators to robotics companies. According to company disclosures, total orders for humanoid robot-related businesses reached approximately 400 million euros in the third quarter of 2026.
Aptiv utilizes existing auto parts factories, while Schaeffler has built new facilities that combine manufacturing and training functions. Under both paths, real production environments have been incorporated into robot product development and industrial cooperation.
This is the practical meaning of "selling scenarios": it's not about directly renting out production lines or selling industrial data, but rather treating real processes, engineering standards, operational feedback, and training verification conditions as cooperative resources—combined with robot procurement, parts supply, and joint development.
In the past, the production line was a guarantee for Tier 1 suppliers to deliver products; today, it has also become a mandatory test that robotics companies must pass to enter the industrial market.
Once Inside the Factory, Robots Face More Hurdles
Connector assembly can be demonstrated repeatedly on a show floor. But in a real factory, completing the action is just the starting point.
Production lines have fixed rhythms, parts have dimensional and positional deviations, and equipment must remain stable over long periods. Robots not only need to finish the job but also meet industrial customers' requirements for quality, efficiency, and safety.
Xu Daquan, President of Bosch China, mentioned a specific engineering challenge during the expo.
Image source: Bosch
When some robots handle heavy objects, their joints and motors heat up continuously, potentially requiring shutdowns for cooling after about an hour of operation. Production tempo, operational precision, and part consistency all face practical challenges as well.
A gap clearly exists between one hour of operational performance and the stable capability needed to cover a full production shift.
These problems cannot be solved by model training alone. Thermal design, joint reliability, mechanical structure, and manufacturing consistency all need to be repeatedly tested under actual working conditions.
Bosch is currently using its own manufacturing system for this very task.
This May, Bosch partnered with Spirit AI, planning to leverage its Chinese factories and logistics centers for robot data collection, model training, and industrial application verification, while also exploring the supply of core components like actuators and sensors.
According to Bosch's disclosed plans, it will push for robot trials at approximately 100 to 200 workstations in China by 2026.
These workstations are both potential jobs for robots and actual testing grounds for their engineering capabilities.
Valeo's practices reflect the process required from technology screening to factory application.
Since June 2022, Valeo has evaluated 43 robot suppliers globally and selected 11 partners for actual testing in 2025. In 2026, the company will further advance the construction of its Nanjing Humanoid Robot Training Center.
Gu Jianmin, CTO of Valeo China, revealed in a previous interview with Gasgoo that the company has launched robot trial operations at its Shenzhen and Nanjing factories, conducting data collection and model training through the training center.
From supplier evaluation to actual production line trial runs, Valeo has undergone years of technical assessment and application testing. Whether a robot can complete movements is just one part of the equation; satisfying industrial production requirements involves long-term stability, process adaptation, and cost of use.
These are precisely the answers that robotics companies find difficult to obtain through laboratory training alone.
Chen Tongqing, co-founder of Tars, stated during the signing ceremony at the expo: "Embodied models define the upper limit of capability, the hardware determines the boundaries of implementation, but trusted performance decides whether physical AI can truly enter the production line."
The collaboration between Aptiv and Tars has already yielded some phased progress.
According to information disclosed by the companies, the first batch of robots has been deployed in actual production scenarios, with relevant robot clusters verified through round-the-clock operation. After about three months of pre-training, the speed of individual actions can basically match that of skilled workers.
From completing assembly actions to catching up to manual operation speeds, training under real working conditions has produced tangible progress. However, single-action speed is not the same metric as overall workstation efficiency, and sustained operational capability still requires further verification.
The value of industrial scenarios is not just about giving robots a place to work.
Real production lines can continuously generate task feedback, helping development teams identify operational deviations, quality issues, and anomalies. However, factory data cannot be directly used in its entirety for model training. Equipment energy consumption, output, and quality records differ from the visual, motion, and contact state data robots require; the latter needs to be collected, processed, and correlated with task results.
Even more difficult is whether the experience a robot accumulates in one factory can be transferred to the next.
Schaeffler's Taicang Digital Intelligence Factory selects typical processes from its global manufacturing system for scenario replication, then verifies training results through digital twins and actual production lines—precisely exploring the connection between training and real-world application.
Tooling, equipment layout, and production rhythms vary between factories. Even when performing the same task, robots may need to readapt.
Successful training on one production line solves the problem for one workstation; stable operation on another line is what turns training results into scalable capability.
Multinational Tier 1 suppliers possess manufacturing networks covering different regions, allowing them to provide diverse working conditions. But the number of factories is just the foundation; whether training results can be reused determines how effective these scenario resources can be.
And this brings the question back to the commercial level.
Scenarios Are on the Table, But What About the Price?
Auto Tier 1 suppliers are starting to offer robotics companies more than just components—they are offering industrial verification and data collection conditions.
Yet if these collaborations are placed on a commercial ledger, one value remains hard to measure: how much is a scenario actually worth?
Schaeffler disclosed total orders of about 400 million euros for humanoid robot business, while Aptiv announced project awards and mass production plans for core technology products.
These are all advances in product supply business. However, to date, no clear scenario usage fees or industrial data sales revenue have been found in these companies' financial reports.
The production line has become a cooperative resource, but scenarios still lack a publicly comparable price list.
Of course, this business doesn't necessarily require separate charges. Robotics companies gain training, verification, and future project opportunities through real production lines, while Tier 1 suppliers may benefit from improved production efficiency, joint development, and core parts matching.
The data-driven production returns mentioned by Yang are one such possible payoff.
The question is whether these returns can cover the added costs of opening scenarios, conducting engineering verification, and deploying robots.
Factories need to coordinate production workstations, dedicate engineering staff, and bear the impact of equipment debugging and operational anomalies. Robotics companies, meanwhile, must continuously optimize models, adjust hardware, and perfect system integration.
Moreover, automotive factories have never been short of automation solutions.
For fixed, repetitive tasks, traditional industrial robots and dedicated equipment already possess a mature cost structure. Embodied robots have a better chance of proving their value in flexible assembly, multi-task switching, and complex operations.
Simple work is already handled well by traditional automation; complex work is where embodied robots are still proving themselves.
Aptiv's plan to deploy hundreds of units has news value, but the actual task success rate, human intervention frequency, maintenance costs, and overall production efficiency during operation will determine whether factories are willing to expand their use.
Scenario data also has usage boundaries.
Process parameters, quality information, and customer data in auto factories may involve trade secrets. Whether robotics companies can use data collected from one factory for other clients, and whether training results can be reused across companies and regions, must be clarified in cooperation mechanisms.
Beyond data boundaries, foreign Tier 1 suppliers face another layer of competition: industrial scenarios are not their exclusive resource.
This February, Joyson Electronics disclosed that humanoid robots have entered its Ningbo automotive safety parts factory to conduct data collection, model training, and industrial scenario adaptation. Joyson is also developing core products like robot controllers.
In October, Yanfeng International signed a cooperation agreement with the National and Local Joint Humanoid Robot Innovation Center, proposing the joint construction of an automotive industry embodied intelligence scenario data center and a training demonstration site.
Local Tier 1 suppliers are similarly incorporating manufacturing resources into robotics industry cooperation.
For multinational Tier 1 suppliers, mature engineering systems and global manufacturing networks provide richer working conditions. Bosch is promoting industrial application verification in China and planning robot contract manufacturing in Germany, while Schaeffler connects its Taicang training facilities with its global manufacturing system.
These layouts provide conditions for cross-factory coordination, but whether they can form replicable solutions in the future still requires verification through actual projects.
As more auto parts companies open their production lines, robotics companies will have more choices. The value of a scenario no longer depends solely on factory scale, but also on what real tasks can be provided, whether valid feedback can be continuously generated, and whether training results can be applied to more production sites.
For Tier 1 suppliers, factories also need to see actual returns from robot applications. Scenario cooperation can help technology move toward mass production, but it could also stall at the verification stage due to high costs or insufficient results.
At this year's expo, Aptiv brought robots onto its production lines and integrated its computing and power management technologies into robot products. Schaeffler, Bosch, and Valeo are advancing similar collaborations at different stages.
A relationship exceeding traditional procurement and supply is forming between auto Tier 1 suppliers and embodied intelligence companies. Production lines not only manufacture products but are also beginning to participate in the development and verification of robot capabilities.
The only question is, when more and more companies are willing to open their factories, why should robotics companies choose your scenario?
Whoever can enable robots to solve more real problems here and carry that experience to the next factory will have the best chance of turning scenarios into a sustainable business.
